{ "model_id": "m5", "model_tag": "gemma4:26b", "arm": "split", "box_class": "desktop", "mode": "context-ladder", "arm_cards": [ "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b", "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962" ], "target_uuid": "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b", "started_utc": "2026-09-17T00:39:09Z", "prompt_sha256": "90eedd0c53f9554ae3837674504fcb7090013d9a432a653352183c0f25a7ce5c", "num_predict": 256, "ladder_note": "The frozen prompt is about 537 tokens. Climbing this ladder measures the cost of RESERVING a context window, NOT the cost of filling one: a 128k window with 537 tokens in it decodes at nearly the speed of an 8k window with 537 tokens in it. What changes is the VRAM the reservation costs, and therefore what fits. Any tok/s difference across levels is a second-order effect of the allocation and is not a long-prompt number. The FILLED-WINDOW arm in this same directory is the one that answers the long-prompt question, and the two must never be quoted as if they were the same measurement.", "fit_rule_note": "A context rung FITS when `/api/ps` reports `size_vram == size` -- every byte of weights and KV on the two cards, nothing in host RAM -- and the ollama runner's own resident memory stays inside the allowance below. A rung that spills is printed as `does not fit`, with its per-card memory, and the ladder stops there: a partial residency is a different measurement and is never reported as the pair's window.", "pcie_note": "The two cards did NOT negotiate the same link. One runs at PCIe x16 and the other at x4, both of a x16-capable generation-3 link -- a property of the SLOT each card landed in on a consumer board, not of the cards. A model split across two cards moves activations between them on every token, so the narrow link is in the path of the measurement and not beside it. Every run records `pcie.link.width.current` on both cards UNDER LOAD, because an idle card drops its link to save power and a width read at rest would flatter the result.", "thermal_layout_note": "Two GeForce RTX 3080 10 GB cards in one consumer desktop, in adjacent slots of the same board, sharing the case's airflow. A temperature here is a reading of THAT arrangement and not of either card in isolation: the upper card sits in the air the lower one has already warmed, and the two are close enough that neither has open air on both faces. Both cards are traced on every run and reported separately, so an asymmetry between them is visible rather than averaged away.", "card_labels": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": "index 0 - Gigabyte - GPU-564e9594", "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": "index 1 - EVGA - GPU-46e03ea9" }, "card_identity": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": 0, "bus_id": "00000000:01:00.0", "pci_sub_device_id": "0x403F1458", "vendor": "Gigabyte", "vbios": "94.02.42.40.33", "pcie_link_width_max": 16, "pcie_link_width_negotiated": 16, "power_default_limit_w": 370.0, "power_max_limit_w": 370.0, "note": "the board that shipped with thermal pads missing -- backplate contact and some memory modules had none -- and was re-padded by hand before this bench. Its heat is a thing this bench was asked to look at directly." }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": 1, "bus_id": "00000000:03:00.0", "pci_sub_device_id": "0x38973842", "vendor": "EVGA", "vbios": "94.02.42.80.16", "pcie_link_width_max": 16, "pcie_link_width_negotiated": 4, "power_default_limit_w": 380.0, "power_max_limit_w": 400.0, "note": "the FTW3 board; the same card the single-3080 bench of 2026-09-10 measured. It is the one in the x4 slot." } }, "memory_temp_support": { "checked_utc": "2026-09-17T00:39:09Z", "paths": { "nvidia-smi --query-gpu=temperature.memory": { "raw": "0, GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b, N/A\n1, GPU-46e03ea9-af32-5a79-21c9-e59c4413e962, N/A", "readable": false }, "nvidia-smi -q -d TEMPERATURE": { "lines": [ "GPU Shutdown Temp : 98 C", "GPU Slowdown Temp : 95 C", "GPU Target Temperature : 83 C", "Memory Current Temp : N/A", "Memory Max Operating Temp : N/A", "GPU Shutdown Temp : 98 C", "GPU Slowdown Temp : 95 C", "GPU Target Temperature : 83 C", "Memory Current Temp : N/A", "Memory Max Operating Temp : N/A" ], "readable": false }, "NVML NVML_FI_DEV_MEMORY_TEMP": { "available": true, "field_id": 82, "cards": { "0": { "call_rc": 0, "field_rc": 3, "supported": false, "not_supported": true, "value_c": null }, "1": { "call_rc": 0, "field_rc": 3, "supported": false, "not_supported": true, "value_c": null } } } }, "readable": false, "verdict": "the memory die's temperature is NOT readable on these cards through any of the three paths asked, so the memory-temperature stop condition could not arm and NO memory temperature is reported anywhere in this bench. The core temperature stop and the driver's own thermal-slowdown reasons are the thermal instrument instead." }, "runner_rss_allowance_gib": 4.0, "num_gpu_option": 99, "num_gpu_policy": "layer count FORCED to 99 via options.num_gpu", "num_gpu_note": "ollama decides for itself how many of a model's layers to put on the cards. On this pair that decision is measurably conservative: with the layer count left to the server, gemma4:26b loaded 75.2% into VRAM and spilled 4.25 GiB to host RAM while about 5 GiB of the two cards' 20 GiB sat unused. Naming the layer count with `options.num_gpu` loaded the SAME model 100% into VRAM. Both are measured and both are reported: `auto` is what a buyer gets out of the box, and the forced count is what the hardware can actually do. A value at or above the model's layer count means every layer; the value sent is recorded in every run record as `num_gpu_option`.", "instance_env": { "written_utc": "2026-09-17T00:32:49Z", "unit": "bench-2x3080-split", "shape": "split", "port": 11470, "base_url": "http://127.0.0.1:11470", "cuda_visible_devices": "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b,GPU-46e03ea9-af32-5a79-21c9-e59c4413e962", "OLLAMA_FLASH_ATTENTION": "1", "OLLAMA_KV_CACHE_TYPE": "q8_0", "OLLAMA_CONTEXT_LENGTH": 32768, "OLLAMA_": 1, "OLLAMA_MAX_LOADED_MODELS": 1, "OLLAMA_KEEP_ALIVE": "0 (the server default; every bench request sends its own keep_alive, which wins, and every arm unloads on the way out)", "models_dir": "/usr/share/ollama/.ollama/models", "models_dir_writable_by_this_user": false, "api_version": { "version": "0.32.13" }, "power_limits_at_start_w": "0, 300.00 W;1, 300.00 W;", "persistence_mode_at_start": "0, Enabled;1, Enabled;", "pcie_link_width_at_start": "0, 16;1, 4;" }, "ups_before_arm": { "ups": "pr1500@localhost", "read_utc": "2026-09-17T00:39:09Z", "available": true, "raw": "battery.charge: 100\nbattery.charge.low: 10\nbattery.charge.warning: 20\nbattery.mfr.date: CPS\nbattery.runtime: 3675\nbattery.runtime.low: 300\nbattery.status: 100%\nbattery.type: PbAcid\nbattery.voltage: 27.5\nbattery.voltage.nominal: 24\ndevice.mfr: CPS\ndevice.model: CP1500PFCLCDa\ndevice.serial: CXXRS7000436\ndevice.type: ups\ndriver.debug: 0\ndriver.flag.allow_killpower: 0\ndriver.name: usbhid-ups\ndriver.parameter.interrupt_pipe_no_events_tolerance: -1\ndriver.parameter.pollfreq: 12\ndriver.parameter.pollinterval: 2\ndriver.parameter.port: auto\ndriver.parameter.synchronous: auto\ndriver.state: quiet\ndriver.version: 2.8.4\ndriver.version.data: CyberPower HID 0.84\ndriver.version.internal: 0.67\ndriver.version.usb: libusb-1.0.29 (API: 0x0100010B)\ninput.sensitivity: normal\ninput.transfer.high: 139\ninput.transfer.low: 100\ninput.voltage: 122.0\ninput.voltage.nominal: 120\noutput.voltage: 122.0\nups.beeper.status: enabled\nups.delay.shutdown: 60\nups.delay.start: 120\nups.firmware: CR01802H1\nups.load: 12\nups.mfr: CPS\nups.model: CP1500PFCLCDa\nups.power: 125\nups.power.nominal: 1500\nups.productid: 0601\nups.realpower: 125\nups.realpower.nominal: 1000\nups.serial: CXXRS7000436\nups.status: OL\nups.test.result: No test initiated\nups.timer.shutdown: -60\nups.timer.start: -60\nups.vendorid: 0764", "battery.charge": 100.0, "input.voltage": 122.0, "output.voltage": 122.0, "ups.load": 12.0, "ups.realpower": 125.0, "ups.realpower.nominal": 1000.0, "ups.status": "OL", "load_pct": 12.0, "stop_threshold_pct": 80.0, "over_stop_threshold": false }, "levels": [ { "num_ctx": 4096, "started_utc": "2026-09-17T00:39:09Z", "loads": true, "size_total": 17238862068, "size_vram": 17238862068, "context_length_loaded": 4096, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8707.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8339.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8960.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8356.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 253.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 17.0 }, "runner_peak_rss_gib": 2.53, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 15489, "rss_kb": 2649372, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 4908 22052 75 20682 42131\nSwap: 24575 0 24575", "kv_footprint_bytes": 0, "kv_footprint_gib": 0.0, "runs": [ { "ttft_ms": 5050.51, "first_token_was_thinking": false, "wall_s": 7.107, "eval_count": 256, "eval_duration_ns": 2053496000, "decode_tok_s": 124.665, "prompt_eval_count": 536, "prompt_eval_duration_ns": 4579992000, "prefill_tok_s": 117.031, "load_duration_ns": 466031560, "total_duration_ns": 7103563222, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.995, "decode_tok_s_wall_gross": 36.021, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 95.35, "median": 95.36, "max": 95.37, "mean": 95.36, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 9, "mean_w": 144.15, "max_w": 191.19, "median_w": 139.34, "mean_w_whole_window": 113.05, "max_w_whole_window": 191.19, "limit_w": 300.0, "over_idle_w": 48.79, "samples": 26, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 95.41, "median": 95.49, "max": 191.19, "mean": 113.05, "n": 26 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 26 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 44.0, "mean": 10.6, "n": 26 }, "temperature_c": { "min": 35.0, "median": 36.0, "max": 42.0, "mean": 37.2, "n": 26 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 26 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 26 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1800.0, "max": 1980.0, "mean": 1845.0, "n": 26 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 26 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 26, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 95.45, "median": 139.34, "max": 191.19, "mean": 144.15, "n": 9 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 9 }, "utilization_pct": { "min": 6.0, "median": 42.0, "max": 44.0, "mean": 30.7, "n": 9 }, "temperature_c": { "min": 36.0, "median": 41.0, "max": 42.0, "mean": 39.4, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1980.0, "max": 1980.0, "mean": 1920.0, "n": 9 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9 }, "busy_samples": 9 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 86.87, "median": 87.14, "max": 207.1, "mean": 110.39, "n": 26 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8356.0, "median": 8356.0, "max": 8356.0, "mean": 8356.0, "n": 26 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 57.0, "mean": 13.0, "n": 26 }, "temperature_c": { "min": 33.0, "median": 34.0, "max": 39.0, "mean": 35.0, "n": 26 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 26 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 26 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1800.0, "max": 1965.0, "mean": 1844.4, "n": 26 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 26 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 26, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 137.49, "median": 202.39, "max": 207.1, "mean": 184.82, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8356.0, "median": 8356.0, "max": 8356.0, "mean": 8356.0, "n": 6 }, "utilization_pct": { "min": 56.0, "median": 56.0, "max": 57.0, "mean": 56.3, "n": 6 }, "temperature_c": { "min": 37.0, "median": 38.0, "max": 39.0, "mean": 38.2, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1156.29, "watts_per_tok_s": 1.1563, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 9, "temp_max_c": 42.0, "temp_min_c": 36.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 144.15, "power_pct_of_limit": 48.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 48.1% of the cap and the card at 42 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 9, "temp_max_c": 42.0, "temp_min_c": 36.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 144.15, "power_pct_of_limit": 48.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 48.1% of the cap and the card at 42 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 39.0, "temp_min_c": 37.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 184.82, "power_pct_of_limit": 61.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8960.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8356.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 26 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 26 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 4, "load_pct": { "min": 21.0, "median": 31.5, "max": 39.0, "mean": 30.8, "n": 4 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 562.24, "first_token_was_thinking": false, "wall_s": 2.6164, "eval_count": 256, "eval_duration_ns": 2051245000, "decode_tok_s": 124.802, "prompt_eval_count": 536, "prompt_eval_duration_ns": 99109000, "prefill_tok_s": 5408.187, "load_duration_ns": 458922207, "total_duration_ns": 2613186907, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.139, "decode_tok_s_wall_gross": 97.845, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 96.27, "median": 96.37, "max": 96.38, "mean": 96.34, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 171.26, "max_w": 192.35, "median_w": 183.87, "mean_w_whole_window": 142.23, "max_w_whole_window": 192.35, "limit_w": 300.0, "over_idle_w": 74.92, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 96.36, "median": 137.78, "max": 192.35, "mean": 142.23, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 40.5, "max": 42.0, "mean": 24.8, "n": 10 }, "temperature_c": { "min": 38.0, "median": 42.0, "max": 44.0, "mean": 41.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1980.0, "max": 1980.0, "mean": 1921.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 131.65, "median": 183.87, "max": 192.35, "mean": 171.26, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 6 }, "utilization_pct": { "min": 40.0, "median": 41.5, "max": 42.0, "mean": 41.3, "n": 6 }, "temperature_c": { "min": 41.0, "median": 43.0, "max": 44.0, "mean": 43.0, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1980.0, "median": 1980.0, "max": 1980.0, "mean": 1980.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 87.28, "median": 144.03, "max": 208.4, "mean": 147.26, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8356.0, "median": 8356.0, "max": 8356.0, "mean": 8356.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 58.0, "mean": 33.9, "n": 10 }, "temperature_c": { "min": 36.0, "median": 39.5, "max": 40.0, "mean": 38.5, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1957.5, "max": 1965.0, "mean": 1912.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 136.45, "median": 203.66, "max": 208.4, "mean": 185.32, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8356.0, "median": 8356.0, "max": 8356.0, "mean": 8356.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 56.5, "max": 58.0, "mean": 56.5, "n": 6 }, "temperature_c": { "min": 39.0, "median": 40.0, "max": 40.0, "mean": 39.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1965.0, "max": 1965.0, "mean": 1960.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1372.25, "watts_per_tok_s": 1.3723, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 44.0, "temp_min_c": 41.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 171.26, "power_pct_of_limit": 57.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 44.0, "temp_min_c": 41.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 171.26, "power_pct_of_limit": 57.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 40.0, "temp_min_c": 39.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 185.32, "power_pct_of_limit": 61.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8960.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8356.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 13.0, "median": 14.0, "max": 15.0, "mean": 14.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 562.77, "first_token_was_thinking": false, "wall_s": 2.6212, "eval_count": 256, "eval_duration_ns": 2055366000, "decode_tok_s": 124.552, "prompt_eval_count": 536, "prompt_eval_duration_ns": 99207000, "prefill_tok_s": 5402.845, "load_duration_ns": 459106254, "total_duration_ns": 2617835820, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.878, "decode_tok_s_wall_gross": 97.664, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 96.75, "median": 96.92, "max": 97.0, "mean": 96.89, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 179.96, "max_w": 192.86, "median_w": 189.16, "mean_w_whole_window": 142.69, "max_w_whole_window": 192.86, "limit_w": 300.0, "over_idle_w": 83.07, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 97.01, "median": 137.05, "max": 192.86, "mean": 142.69, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 20.5, "max": 43.0, "mean": 21.0, "n": 10 }, "temperature_c": { "min": 39.0, "median": 44.0, "max": 45.0, "mean": 42.6, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1980.0, "mean": 1912.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 143.35, "median": 189.16, "max": 192.86, "mean": 179.96, "n": 5 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8960.0, "median": 8960.0, "max": 8960.0, "mean": 8960.0, "n": 5 }, "utilization_pct": { "min": 41.0, "median": 42.0, "max": 43.0, "mean": 42.0, "n": 5 }, "temperature_c": { "min": 44.0, "median": 45.0, "max": 45.0, "mean": 44.8, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 5 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 88.17, "median": 146.06, "max": 208.62, "mean": 148.53, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8356.0, "median": 8356.0, "max": 8356.0, "mean": 8356.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 27.5, "max": 57.0, "mean": 28.1, "n": 10 }, "temperature_c": { "min": 38.0, "median": 41.5, "max": 42.0, "mean": 40.5, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 150.82, "median": 207.96, "max": 208.62, "mean": 195.21, "n": 5 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8356.0, "median": 8356.0, "max": 8356.0, "mean": 8356.0, "n": 5 }, "utilization_pct": { "min": 55.0, "median": 57.0, "max": 57.0, "mean": 56.2, "n": 5 }, "temperature_c": { "min": 42.0, "median": 42.0, "max": 42.0, "mean": 42.0, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 5 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 5 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1444.86, "watts_per_tok_s": 1.4449, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 45.0, "temp_min_c": 44.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 179.96, "power_pct_of_limit": 60.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 45.0, "temp_min_c": 44.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 179.96, "power_pct_of_limit": 60.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 42.0, "temp_min_c": 42.0, "temp_rise_c": 0.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 195.21, "power_pct_of_limit": 65.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8960.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8356.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 124.552, "median": 124.665, "max": 124.802, "mean": 124.673, "n": 3 }, "ttft_ms": { "min": 562.24, "median": 562.77, "max": 5050.51, "mean": 2058.51, "n": 3 }, "power_mean_w": { "min": 144.15, "median": 171.26, "max": 179.96, "mean": 165.12, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1156.29, "median": 1372.25, "max": 1444.86, "mean": 1324.47, "n": 3 } }, "finished_utc": "2026-09-17T00:40:15Z" }, { "num_ctx": 8192, "started_utc": "2026-09-17T00:40:15Z", "loads": true, "size_total": 17272416500, "size_vram": 17272416500, "context_length_loaded": 8192, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 32.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 42.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8992.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8398.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8960.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8356.0 }, "runner_peak_rss_gib": 2.54, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 16491, "rss_kb": 2666092, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 5109 21836 91 20713 41930\nSwap: 24575 0 24575", "kv_footprint_bytes": 33554432, "kv_footprint_gib": 0.03, "runs": [ { "ttft_ms": 586.43, "first_token_was_thinking": false, "wall_s": 2.6491, "eval_count": 256, "eval_duration_ns": 2059580000, "decode_tok_s": 124.297, "prompt_eval_count": 536, "prompt_eval_duration_ns": 114402000, "prefill_tok_s": 4685.233, "load_duration_ns": 467725428, "total_duration_ns": 2645729686, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.624, "decode_tok_s_wall_gross": 96.635, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 99.08, "median": 99.08, "max": 99.14, "mean": 99.1, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 171.38, "max_w": 194.07, "median_w": 186.16, "mean_w_whole_window": 143.28, "max_w_whole_window": 194.07, "limit_w": 300.0, "over_idle_w": 72.28, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 99.18, "median": 133.91, "max": 194.07, "mean": 143.28, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 42.0, "max": 43.0, "mean": 25.6, "n": 10 }, "temperature_c": { "min": 42.0, "median": 46.5, "max": 48.0, "mean": 45.5, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1957.5, "max": 1965.0, "mean": 1900.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 127.35, "median": 186.16, "max": 194.07, "mean": 171.38, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 6 }, "utilization_pct": { "min": 41.0, "median": 43.0, "max": 43.0, "mean": 42.7, "n": 6 }, "temperature_c": { "min": 46.0, "median": 48.0, "max": 48.0, "mean": 47.5, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1965.0, "max": 1965.0, "mean": 1962.5, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 88.58, "median": 145.75, "max": 210.79, "mean": 149.52, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8398.0, "median": 8398.0, "max": 8398.0, "mean": 8398.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 58.0, "mean": 33.7, "n": 10 }, "temperature_c": { "min": 40.0, "median": 44.5, "max": 45.0, "mean": 43.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 138.5, "median": 207.75, "max": 210.79, "mean": 188.04, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8398.0, "median": 8398.0, "max": 8398.0, "mean": 8398.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 55.5, "max": 58.0, "mean": 56.2, "n": 6 }, "temperature_c": { "min": 44.0, "median": 45.0, "max": 45.0, "mean": 44.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1378.79, "watts_per_tok_s": 1.3788, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 48.0, "temp_min_c": 46.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 171.38, "power_pct_of_limit": 57.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 48.0, "temp_min_c": 46.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 171.38, "power_pct_of_limit": 57.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 45.0, "temp_min_c": 44.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 188.04, "power_pct_of_limit": 62.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8992.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8398.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 571.8, "first_token_was_thinking": false, "wall_s": 2.6336, "eval_count": 256, "eval_duration_ns": 2058743000, "decode_tok_s": 124.348, "prompt_eval_count": 536, "prompt_eval_duration_ns": 99798000, "prefill_tok_s": 5370.849, "load_duration_ns": 467559410, "total_duration_ns": 2630216007, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.681, "decode_tok_s_wall_gross": 97.207, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 99.32, "median": 99.4, "max": 99.48, "mean": 99.4, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 162.43, "max_w": 194.9, "median_w": 176.39, "mean_w_whole_window": 143.57, "max_w_whole_window": 194.9, "limit_w": 300.0, "over_idle_w": 63.03, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 99.39, "median": 137.25, "max": 194.9, "mean": 143.57, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 42.0, "max": 44.0, "mean": 29.8, "n": 10 }, "temperature_c": { "min": 44.0, "median": 48.0, "max": 50.0, "mean": 47.2, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1957.5, "max": 1965.0, "mean": 1897.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 108.26, "median": 176.39, "max": 194.9, "mean": 162.43, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 7 }, "utilization_pct": { "min": 42.0, "median": 42.0, "max": 44.0, "mean": 42.6, "n": 7 }, "temperature_c": { "min": 46.0, "median": 49.0, "max": 50.0, "mean": 48.6, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1939.3, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 88.91, "median": 148.44, "max": 212.09, "mean": 151.28, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8398.0, "median": 8398.0, "max": 8398.0, "mean": 8398.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 58.0, "mean": 39.3, "n": 10 }, "temperature_c": { "min": 42.0, "median": 46.0, "max": 47.0, "mean": 44.8, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1950.0, "mean": 1896.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.37, "median": 206.33, "max": 212.09, "mean": 177.86, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8398.0, "median": 8398.0, "max": 8398.0, "mean": 8398.0, "n": 7 }, "utilization_pct": { "min": 55.0, "median": 56.0, "max": 58.0, "mean": 56.1, "n": 7 }, "temperature_c": { "min": 45.0, "median": 46.0, "max": 47.0, "mean": 46.0, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1950.0, "mean": 1937.1, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1306.26, "watts_per_tok_s": 1.3063, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 50.0, "temp_min_c": 46.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 162.43, "power_pct_of_limit": 54.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 54.1% of the cap and the card at 50 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 50.0, "temp_min_c": 46.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 162.43, "power_pct_of_limit": 54.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 54.1% of the cap and the card at 50 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 47.0, "temp_min_c": 45.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1935.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 177.86, "power_pct_of_limit": 59.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8992.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8398.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 13.0, "median": 13.0, "max": 13.0, "mean": 13.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 606.23, "first_token_was_thinking": false, "wall_s": 2.6689, "eval_count": 256, "eval_duration_ns": 2059623999, "decode_tok_s": 124.295, "prompt_eval_count": 536, "prompt_eval_duration_ns": 141338000, "prefill_tok_s": 3792.328, "load_duration_ns": 460660710, "total_duration_ns": 2665679589, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.624, "decode_tok_s_wall_gross": 95.918, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 15.92, "median": 15.93, "max": 16.16, "mean": 16.0, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 145.57, "max_w": 195.41, "median_w": 179.21, "mean_w_whole_window": 107.38, "max_w_whole_window": 195.41, "limit_w": 300.0, "over_idle_w": 129.57, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 16.38, "median": 107.78, "max": 195.41, "mean": 107.38, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 11.0, "max": 42.0, "mean": 19.9, "n": 10 }, "temperature_c": { "min": 40.0, "median": 48.0, "max": 50.0, "mean": 46.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1965.0, "mean": 1585.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 42.83, "median": 179.21, "max": 195.41, "mean": 145.57, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8992.0, "median": 8992.0, "max": 8992.0, "mean": 8992.0, "n": 7 }, "utilization_pct": { "min": 11.0, "median": 41.0, "max": 42.0, "mean": 28.4, "n": 7 }, "temperature_c": { "min": 46.0, "median": 49.0, "max": 50.0, "mean": 48.7, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1965.0, "max": 1965.0, "mean": 1947.9, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.05, "median": 108.59, "max": 211.86, "mean": 112.98, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8398.0, "median": 8398.0, "max": 8398.0, "mean": 8398.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 38.0, "max": 58.0, "mean": 34.2, "n": 10 }, "temperature_c": { "min": 40.0, "median": 45.5, "max": 47.0, "mean": 44.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1950.0, "mean": 1578.0, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 34.31, "median": 202.83, "max": 211.86, "mean": 155.61, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8398.0, "median": 8398.0, "max": 8398.0, "mean": 8398.0, "n": 7 }, "utilization_pct": { "min": 38.0, "median": 56.0, "max": 58.0, "mean": 48.9, "n": 7 }, "temperature_c": { "min": 44.0, "median": 46.0, "max": 47.0, "mean": 46.0, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1950.0, "mean": 1937.1, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1171.17, "watts_per_tok_s": 1.1712, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 50.0, "temp_min_c": 46.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 145.57, "power_pct_of_limit": 48.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 50.0, "temp_min_c": 46.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 145.57, "power_pct_of_limit": 48.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 47.0, "temp_min_c": 44.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1935.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 155.61, "power_pct_of_limit": 51.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8992.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8398.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 124.295, "median": 124.297, "max": 124.348, "mean": 124.313, "n": 3 }, "ttft_ms": { "min": 571.8, "median": 586.43, "max": 606.23, "mean": 588.15, "n": 3 }, "power_mean_w": { "min": 145.57, "median": 162.43, "max": 171.38, "mean": 159.79, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1171.17, "median": 1306.26, "max": 1378.79, "mean": 1285.41, "n": 3 } }, "finished_utc": "2026-09-17T00:41:26Z" }, { "num_ctx": 16384, "started_utc": "2026-09-17T00:41:26Z", "loads": true, "size_total": 17399545854, "size_vram": 17399545854, "context_length_loaded": 16384, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 96.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 112.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9088.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8510.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8992.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8398.0 }, "runner_peak_rss_gib": 2.57, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 17377, "rss_kb": 2699180, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 5037 21881 123 20773 42002\nSwap: 24575 0 24575", "kv_footprint_bytes": 160683786, "kv_footprint_gib": 0.15, "runs": [ { "ttft_ms": 581.88, "first_token_was_thinking": false, "wall_s": 2.6575, "eval_count": 256, "eval_duration_ns": 2072510000, "decode_tok_s": 123.522, "prompt_eval_count": 536, "prompt_eval_duration_ns": 114976000, "prefill_tok_s": 4661.842, "load_duration_ns": 462512480, "total_duration_ns": 2654112780, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 122.856, "decode_tok_s_wall_gross": 96.332, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 101.7, "median": 101.73, "max": 101.73, "mean": 101.72, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 164.29, "max_w": 197.12, "median_w": 180.48, "mean_w_whole_window": 145.57, "max_w_whole_window": 197.12, "limit_w": 300.0, "over_idle_w": 62.57, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 101.74, "median": 138.19, "max": 197.12, "mean": 145.57, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 13.0, "max": 43.0, "mean": 20.9, "n": 10 }, "temperature_c": { "min": 47.0, "median": 50.5, "max": 53.0, "mean": 50.1, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1890.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 110.25, "median": 180.48, "max": 197.12, "mean": 164.29, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 7 }, "utilization_pct": { "min": 13.0, "median": 42.0, "max": 43.0, "mean": 29.9, "n": 7 }, "temperature_c": { "min": 49.0, "median": 52.0, "max": 53.0, "mean": 51.4, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1928.6, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.17, "median": 138.88, "max": 214.24, "mean": 148.98, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8510.0, "median": 8510.0, "max": 8510.0, "mean": 8510.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 28.0, "max": 58.0, "mean": 31.0, "n": 10 }, "temperature_c": { "min": 45.0, "median": 49.0, "max": 50.0, "mean": 47.8, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1935.0, "mean": 1881.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 102.09, "median": 201.19, "max": 214.24, "mean": 174.5, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8510.0, "median": 8510.0, "max": 8510.0, "mean": 8510.0, "n": 7 }, "utilization_pct": { "min": 28.0, "median": 55.0, "max": 58.0, "mean": 44.3, "n": 7 }, "temperature_c": { "min": 47.0, "median": 49.0, "max": 50.0, "mean": 49.0, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1935.0, "mean": 1915.7, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1330.05, "watts_per_tok_s": 1.33, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 53.0, "temp_min_c": 49.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 164.29, "power_pct_of_limit": 54.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 54.8% of the cap and the card at 53 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 53.0, "temp_min_c": 49.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 164.29, "power_pct_of_limit": 54.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 54.8% of the cap and the card at 53 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 50.0, "temp_min_c": 47.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1935.0, "clock_median_mhz": 1935.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 174.5, "power_pct_of_limit": 58.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 58.2% of the cap and the card at 50 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9088.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8510.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 605.99, "first_token_was_thinking": false, "wall_s": 2.6758, "eval_count": 256, "eval_duration_ns": 2066786999, "decode_tok_s": 123.864, "prompt_eval_count": 536, "prompt_eval_duration_ns": 140507000, "prefill_tok_s": 3814.757, "load_duration_ns": 461199637, "total_duration_ns": 2672409476, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.202, "decode_tok_s_wall_gross": 95.674, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 15.71, "median": 15.73, "max": 16.54, "mean": 15.99, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 164.42, "max_w": 197.84, "median_w": 189.0, "mean_w_whole_window": 107.96, "max_w_whole_window": 197.84, "limit_w": 300.0, "over_idle_w": 148.43, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 15.91, "median": 106.53, "max": 197.84, "mean": 107.96, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 40.5, "max": 42.0, "mean": 24.6, "n": 10 }, "temperature_c": { "min": 43.0, "median": 51.5, "max": 53.0, "mean": 49.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1980.0, "max": 1980.0, "mean": 1597.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 96.55, "median": 189.0, "max": 197.84, "mean": 164.42, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 6 }, "utilization_pct": { "min": 40.0, "median": 41.0, "max": 42.0, "mean": 41.0, "n": 6 }, "temperature_c": { "min": 51.0, "median": 52.5, "max": 53.0, "mean": 52.3, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1980.0, "median": 1980.0, "max": 1980.0, "mean": 1980.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.13, "median": 107.52, "max": 213.73, "mean": 114.16, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8510.0, "median": 8510.0, "max": 8510.0, "mean": 8510.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 58.0, "mean": 33.7, "n": 10 }, "temperature_c": { "min": 43.0, "median": 49.0, "max": 50.0, "mean": 47.2, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1935.0, "mean": 1576.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 95.65, "median": 208.89, "max": 213.73, "mean": 176.68, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8510.0, "median": 8510.0, "max": 8510.0, "mean": 8510.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 55.5, "max": 58.0, "mean": 56.2, "n": 6 }, "temperature_c": { "min": 49.0, "median": 49.0, "max": 50.0, "mean": 49.3, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1935.0, "mean": 1935.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1327.43, "watts_per_tok_s": 1.3274, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 53.0, "temp_min_c": 51.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 164.42, "power_pct_of_limit": 54.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 53.0, "temp_min_c": 51.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 164.42, "power_pct_of_limit": 54.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 50.0, "temp_min_c": 49.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1935.0, "clock_median_mhz": 1935.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 176.68, "power_pct_of_limit": 58.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9088.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8510.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 565.35, "first_token_was_thinking": false, "wall_s": 2.627, "eval_count": 256, "eval_duration_ns": 2058841000, "decode_tok_s": 124.342, "prompt_eval_count": 536, "prompt_eval_duration_ns": 100795000, "prefill_tok_s": 5317.724, "load_duration_ns": 460129013, "total_duration_ns": 2623801436, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.686, "decode_tok_s_wall_gross": 97.449, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 100.56, "median": 100.63, "max": 100.64, "mean": 100.61, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 167.15, "max_w": 199.0, "median_w": 185.15, "mean_w_whole_window": 147.35, "max_w_whole_window": 199.0, "limit_w": 300.0, "over_idle_w": 66.54, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 100.58, "median": 141.27, "max": 199.0, "mean": 147.35, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 40.5, "max": 43.0, "mean": 29.1, "n": 10 }, "temperature_c": { "min": 49.0, "median": 53.5, "max": 55.0, "mean": 52.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1980.0, "max": 1980.0, "mean": 1915.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 108.29, "median": 185.15, "max": 199.0, "mean": 167.15, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9088.0, "median": 9088.0, "max": 9088.0, "mean": 9088.0, "n": 7 }, "utilization_pct": { "min": 40.0, "median": 42.0, "max": 43.0, "mean": 41.6, "n": 7 }, "temperature_c": { "min": 51.0, "median": 54.0, "max": 55.0, "mean": 53.7, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1980.0, "max": 1980.0, "mean": 1965.0, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.43, "median": 150.07, "max": 215.63, "mean": 152.82, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8510.0, "median": 8510.0, "max": 8510.0, "mean": 8510.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 28.0, "max": 57.0, "mean": 28.3, "n": 10 }, "temperature_c": { "min": 47.0, "median": 50.0, "max": 51.0, "mean": 49.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1965.0, "mean": 1909.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 157.4, "median": 214.78, "max": 215.63, "mean": 202.67, "n": 5 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8510.0, "median": 8510.0, "max": 8510.0, "mean": 8510.0, "n": 5 }, "utilization_pct": { "min": 56.0, "median": 57.0, "max": 57.0, "mean": 56.6, "n": 5 }, "temperature_c": { "min": 50.0, "median": 51.0, "max": 51.0, "mean": 50.8, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 5 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 5 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1344.28, "watts_per_tok_s": 1.3443, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 55.0, "temp_min_c": 51.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 167.15, "power_pct_of_limit": 55.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 55.7% of the cap and the card at 55 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 55.0, "temp_min_c": 51.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 167.15, "power_pct_of_limit": 55.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 55.7% of the cap and the card at 55 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 51.0, "temp_min_c": 50.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 202.67, "power_pct_of_limit": 67.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9088.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8510.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 123.522, "median": 123.864, "max": 124.342, "mean": 123.909, "n": 3 }, "ttft_ms": { "min": 565.35, "median": 581.88, "max": 605.99, "mean": 584.41, "n": 3 }, "power_mean_w": { "min": 164.29, "median": 164.42, "max": 167.15, "mean": 165.29, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1327.43, "median": 1330.05, "max": 1344.28, "mean": 1333.92, "n": 3 } }, "finished_utc": "2026-09-17T00:42:38Z" }, { "num_ctx": 32768, "started_utc": "2026-09-17T00:42:38Z", "loads": true, "size_total": 17667981310, "size_vram": 17667981310, "context_length_loaded": 32768, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 196.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 230.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9284.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8740.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9088.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8510.0 }, "runner_peak_rss_gib": 2.64, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 18375, "rss_kb": 2765460, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 4982 21935 187 20837 42057\nSwap: 24575 0 24575", "kv_footprint_bytes": 429119242, "kv_footprint_gib": 0.4, "runs": [ { "ttft_ms": 577.69, "first_token_was_thinking": false, "wall_s": 2.6336, "eval_count": 256, "eval_duration_ns": 2053181000, "decode_tok_s": 124.685, "prompt_eval_count": 536, "prompt_eval_duration_ns": 113623000, "prefill_tok_s": 4717.355, "load_duration_ns": 459683231, "total_duration_ns": 2630557061, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.03, "decode_tok_s_wall_gross": 97.204, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 102.25, "median": 102.29, "max": 102.34, "mean": 102.29, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 167.99, "max_w": 202.37, "median_w": 182.5, "mean_w_whole_window": 148.34, "max_w_whole_window": 202.37, "limit_w": 300.0, "over_idle_w": 65.7, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 102.37, "median": 137.44, "max": 202.37, "mean": 148.34, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 42.0, "max": 43.0, "mean": 29.7, "n": 10 }, "temperature_c": { "min": 50.0, "median": 53.5, "max": 56.0, "mean": 53.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 57.0, "median": 57.0, "max": 57.0, "mean": 57.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1980.0, "mean": 1911.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 118.99, "median": 182.5, "max": 202.37, "mean": 167.99, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 7 }, "utilization_pct": { "min": 42.0, "median": 42.0, "max": 43.0, "mean": 42.4, "n": 7 }, "temperature_c": { "min": 52.0, "median": 55.0, "max": 56.0, "mean": 54.3, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 57.0, "median": 57.0, "max": 57.0, "mean": 57.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1965.0, "max": 1980.0, "mean": 1958.6, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.29, "median": 143.8, "max": 216.55, "mean": 151.15, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8740.0, "median": 8740.0, "max": 8740.0, "mean": 8740.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 56.5, "max": 58.0, "mean": 39.9, "n": 10 }, "temperature_c": { "min": 48.0, "median": 51.5, "max": 53.0, "mean": 50.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1915.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 99.0, "median": 206.73, "max": 216.55, "mean": 177.45, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8740.0, "median": 8740.0, "max": 8740.0, "mean": 8740.0, "n": 7 }, "utilization_pct": { "min": 56.0, "median": 57.0, "max": 58.0, "mean": 57.0, "n": 7 }, "temperature_c": { "min": 50.0, "median": 52.0, "max": 53.0, "mean": 51.9, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1347.32, "watts_per_tok_s": 1.3473, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 56.0, "temp_min_c": 52.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 57.0, "fan_max_pct": 57.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 167.99, "power_pct_of_limit": 56.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 56.0% of the cap and the card at 56 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 56.0, "temp_min_c": 52.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 57.0, "fan_max_pct": 57.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 167.99, "power_pct_of_limit": 56.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 56.0% of the cap and the card at 56 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 53.0, "temp_min_c": 50.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 177.45, "power_pct_of_limit": 59.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9284.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8740.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 607.01, "first_token_was_thinking": false, "wall_s": 2.662, "eval_count": 256, "eval_duration_ns": 2052016000, "decode_tok_s": 124.755, "prompt_eval_count": 536, "prompt_eval_duration_ns": 140119000, "prefill_tok_s": 3825.32, "load_duration_ns": 462250433, "total_duration_ns": 2658651038, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.086, "decode_tok_s_wall_gross": 96.167, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.22, "median": 16.38, "max": 17.83, "mean": 16.81, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 150.93, "max_w": 201.33, "median_w": 185.79, "mean_w_whole_window": 111.24, "max_w_whole_window": 201.33, "limit_w": 300.0, "over_idle_w": 134.12, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 16.45, "median": 100.5, "max": 201.33, "mean": 111.24, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 18.0, "max": 43.0, "mean": 22.4, "n": 10 }, "temperature_c": { "min": 45.0, "median": 53.5, "max": 55.0, "mean": 51.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1980.0, "max": 1980.0, "mean": 1591.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.2, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.2, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 70.12, "median": 185.79, "max": 201.33, "mean": 150.93, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 7 }, "utilization_pct": { "min": 18.0, "median": 42.0, "max": 43.0, "mean": 32.0, "n": 7 }, "temperature_c": { "min": 50.0, "median": 54.0, "max": 55.0, "mean": 53.7, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1815.0, "median": 1980.0, "max": 1980.0, "mean": 1956.4, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.2857, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.2857, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 12.05, "median": 115.95, "max": 216.31, "mean": 116.8, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8740.0, "median": 8740.0, "max": 8740.0, "mean": 8740.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 56.0, "max": 58.0, "mean": 34.0, "n": 10 }, "temperature_c": { "min": 45.0, "median": 51.5, "max": 52.0, "mean": 49.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1965.0, "max": 1965.0, "mean": 1594.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.2, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.2, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 104.74, "median": 213.21, "max": 216.31, "mean": 181.76, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8740.0, "median": 8740.0, "max": 8740.0, "mean": 8740.0, "n": 6 }, "utilization_pct": { "min": 56.0, "median": 56.0, "max": 58.0, "mean": 56.7, "n": 6 }, "temperature_c": { "min": 51.0, "median": 52.0, "max": 52.0, "mean": 51.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.3333, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.3333, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1209.81, "watts_per_tok_s": 1.2098, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 55.0, "temp_min_c": 50.0, "temp_rise_c": 5.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 58.0, "fan_max_pct": 58.0, "clock_floor_mhz": 1815.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 150.93, "power_pct_of_limit": 50.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.2857, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.2857, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 50.3% of the cap and the card at 55 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 55.0, "temp_min_c": 50.0, "temp_rise_c": 5.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 58.0, "fan_max_pct": 58.0, "clock_floor_mhz": 1815.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 150.93, "power_pct_of_limit": 50.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.2857, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.2857, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 50.3% of the cap and the card at 55 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 52.0, "temp_min_c": 51.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 181.76, "power_pct_of_limit": 60.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.3333, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.3333, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9284.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8740.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 603.11, "first_token_was_thinking": false, "wall_s": 2.6652, "eval_count": 256, "eval_duration_ns": 2059030000, "decode_tok_s": 124.33, "prompt_eval_count": 536, "prompt_eval_duration_ns": 139853000, "prefill_tok_s": 3832.596, "load_duration_ns": 458747762, "total_duration_ns": 2661723461, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.658, "decode_tok_s_wall_gross": 96.051, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.4, "median": 16.61, "max": 16.66, "mean": 16.56, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 164.88, "max_w": 200.77, "median_w": 191.74, "mean_w_whole_window": 108.64, "max_w_whole_window": 200.77, "limit_w": 300.0, "over_idle_w": 148.32, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 16.3, "median": 102.16, "max": 200.77, "mean": 108.64, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 41.0, "max": 43.0, "mean": 24.9, "n": 10 }, "temperature_c": { "min": 45.0, "median": 53.0, "max": 55.0, "mean": 51.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1972.5, "max": 1980.0, "mean": 1588.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.3, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.1, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 91.41, "median": 191.74, "max": 200.77, "mean": 164.88, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9284.0, "median": 9284.0, "max": 9284.0, "mean": 9284.0, "n": 6 }, "utilization_pct": { "min": 41.0, "median": 41.0, "max": 43.0, "mean": 41.5, "n": 6 }, "temperature_c": { "min": 52.0, "median": 54.0, "max": 55.0, "mean": 54.0, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1980.0, "max": 1980.0, "mean": 1977.5, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.1667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.1667, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 12.23, "median": 116.99, "max": 216.66, "mean": 117.09, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8740.0, "median": 8740.0, "max": 8740.0, "mean": 8740.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.5, "max": 58.0, "mean": 33.9, "n": 10 }, "temperature_c": { "min": 46.0, "median": 51.5, "max": 52.0, "mean": 50.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1965.0, "max": 1965.0, "mean": 1594.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.1, "median": 211.81, "max": 216.66, "mean": 181.78, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8740.0, "median": 8740.0, "max": 8740.0, "mean": 8740.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 56.0, "max": 58.0, "mean": 56.5, "n": 6 }, "temperature_c": { "min": 51.0, "median": 52.0, "max": 52.0, "mean": 51.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1326.14, "watts_per_tok_s": 1.3261, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 55.0, "temp_min_c": 52.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 58.0, "fan_max_pct": 58.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 164.88, "power_pct_of_limit": 55.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.1667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1667, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 55.0, "temp_min_c": 52.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 58.0, "fan_max_pct": 58.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 164.88, "power_pct_of_limit": 55.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.1667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1667, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 52.0, "temp_min_c": 51.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 181.78, "power_pct_of_limit": 60.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9284.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8740.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 124.33, "median": 124.685, "max": 124.755, "mean": 124.59, "n": 3 }, "ttft_ms": { "min": 577.69, "median": 603.11, "max": 607.01, "mean": 595.94, "n": 3 }, "power_mean_w": { "min": 150.93, "median": 164.88, "max": 167.99, "mean": 161.27, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1209.81, "median": 1326.14, "max": 1347.32, "mean": 1294.42, "n": 3 } }, "finished_utc": "2026-09-17T00:44:00Z" }, { "num_ctx": 49152, "started_utc": "2026-09-17T00:44:00Z", "loads": true, "size_total": 17936416766, "size_vram": 17936416766, "context_length_loaded": 49152, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 196.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 230.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9480.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8970.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9284.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8740.0 }, "runner_peak_rss_gib": 2.7, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 19415, "rss_kb": 2833172, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 5052 21864 251 20902 41987\nSwap: 24575 0 24575", "kv_footprint_bytes": 697554698, "kv_footprint_gib": 0.65, "runs": [ { "ttft_ms": 575.17, "first_token_was_thinking": false, "wall_s": 2.6323, "eval_count": 256, "eval_duration_ns": 2054199999, "decode_tok_s": 124.623, "prompt_eval_count": 536, "prompt_eval_duration_ns": 113260000, "prefill_tok_s": 4732.474, "load_duration_ns": 457595170, "total_duration_ns": 2629042884, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.958, "decode_tok_s_wall_gross": 97.253, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 102.79, "median": 102.82, "max": 102.82, "mean": 102.81, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 176.19, "max_w": 202.1, "median_w": 188.29, "mean_w_whole_window": 147.64, "max_w_whole_window": 202.1, "limit_w": 300.0, "over_idle_w": 73.38, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 102.87, "median": 138.18, "max": 202.1, "mean": 147.64, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 42.0, "max": 43.0, "mean": 25.3, "n": 10 }, "temperature_c": { "min": 51.0, "median": 55.0, "max": 57.0, "mean": 54.2, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1980.0, "max": 1980.0, "mean": 1908.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 131.13, "median": 188.29, "max": 202.1, "mean": 176.19, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 6 }, "utilization_pct": { "min": 42.0, "median": 42.0, "max": 43.0, "mean": 42.2, "n": 6 }, "temperature_c": { "min": 54.0, "median": 56.0, "max": 57.0, "mean": 56.0, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1980.0, "median": 1980.0, "max": 1980.0, "mean": 1980.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.62, "median": 144.56, "max": 217.98, "mean": 152.06, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8970.0, "median": 8970.0, "max": 8970.0, "mean": 8970.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 57.0, "mean": 33.2, "n": 10 }, "temperature_c": { "min": 50.0, "median": 53.0, "max": 54.0, "mean": 52.3, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1915.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 139.36, "median": 211.77, "max": 217.98, "mean": 191.41, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8970.0, "median": 8970.0, "max": 8970.0, "mean": 8970.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 55.0, "max": 57.0, "mean": 55.3, "n": 6 }, "temperature_c": { "min": 53.0, "median": 54.0, "max": 54.0, "mean": 53.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1413.79, "watts_per_tok_s": 1.4138, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 57.0, "temp_min_c": 54.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 58.0, "fan_max_pct": 58.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 176.19, "power_pct_of_limit": 58.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 57.0, "temp_min_c": 54.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 58.0, "fan_max_pct": 58.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 176.19, "power_pct_of_limit": 58.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 54.0, "temp_min_c": 53.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 191.41, "power_pct_of_limit": 63.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9480.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8970.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 13.0, "median": 13.0, "max": 13.0, "mean": 13.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 562.11, "first_token_was_thinking": false, "wall_s": 2.618, "eval_count": 256, "eval_duration_ns": 2052888000, "decode_tok_s": 124.702, "prompt_eval_count": 536, "prompt_eval_duration_ns": 99673000, "prefill_tok_s": 5377.585, "load_duration_ns": 458035162, "total_duration_ns": 2614641409, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.035, "decode_tok_s_wall_gross": 97.785, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 103.49, "median": 103.56, "max": 103.58, "mean": 103.54, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 169.38, "max_w": 203.14, "median_w": 186.41, "mean_w_whole_window": 149.77, "max_w_whole_window": 203.14, "limit_w": 300.0, "over_idle_w": 65.84, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 103.53, "median": 141.75, "max": 203.14, "mean": 149.77, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 40.5, "max": 43.0, "mean": 28.9, "n": 10 }, "temperature_c": { "min": 51.0, "median": 55.5, "max": 57.0, "mean": 54.5, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 59.0, "median": 60.0, "max": 60.0, "mean": 59.9, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1980.0, "max": 1980.0, "mean": 1908.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 111.05, "median": 186.41, "max": 203.14, "mean": 169.38, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 7 }, "utilization_pct": { "min": 40.0, "median": 41.0, "max": 43.0, "mean": 41.3, "n": 7 }, "temperature_c": { "min": 53.0, "median": 57.0, "max": 57.0, "mean": 56.0, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 59.0, "median": 60.0, "max": 60.0, "mean": 59.9, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1980.0, "max": 1980.0, "mean": 1954.3, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.83, "median": 148.01, "max": 219.01, "mean": 153.33, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8970.0, "median": 8970.0, "max": 8970.0, "mean": 8970.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 56.0, "max": 56.0, "mean": 39.2, "n": 10 }, "temperature_c": { "min": 50.0, "median": 54.0, "max": 54.0, "mean": 52.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1915.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 101.7, "median": 209.7, "max": 219.01, "mean": 180.5, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8970.0, "median": 8970.0, "max": 8970.0, "mean": 8970.0, "n": 7 }, "utilization_pct": { "min": 56.0, "median": 56.0, "max": 56.0, "mean": 56.0, "n": 7 }, "temperature_c": { "min": 52.0, "median": 54.0, "max": 54.0, "mean": 53.7, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1358.27, "watts_per_tok_s": 1.3583, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 57.0, "temp_min_c": 53.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 59.9, "fan_max_pct": 60.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 169.38, "power_pct_of_limit": 56.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 56.5% of the cap and the card at 57 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 57.0, "temp_min_c": 53.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 59.9, "fan_max_pct": 60.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 169.38, "power_pct_of_limit": 56.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 56.5% of the cap and the card at 57 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 54.0, "temp_min_c": 52.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 180.5, "power_pct_of_limit": 60.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9480.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8970.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 611.99, "first_token_was_thinking": false, "wall_s": 2.6654, "eval_count": 256, "eval_duration_ns": 2050610999, "decode_tok_s": 124.841, "prompt_eval_count": 536, "prompt_eval_duration_ns": 140075000, "prefill_tok_s": 3826.522, "load_duration_ns": 467340488, "total_duration_ns": 2662186738, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.182, "decode_tok_s_wall_gross": 96.045, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.62, "median": 16.92, "max": 17.16, "mean": 16.9, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 149.63, "max_w": 202.37, "median_w": 189.32, "mean_w_whole_window": 109.93, "max_w_whole_window": 202.37, "limit_w": 300.0, "over_idle_w": 132.73, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 16.33, "median": 106.81, "max": 202.37, "mean": 109.93, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 6.0, "max": 43.0, "mean": 18.8, "n": 10 }, "temperature_c": { "min": 47.0, "median": 54.5, "max": 56.0, "mean": 52.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 60.0, "median": 60.0, "max": 60.0, "mean": 60.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1980.0, "max": 1980.0, "mean": 1599.0, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.2, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.2, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 40.36, "median": 189.32, "max": 202.37, "mean": 149.63, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9480.0, "median": 9480.0, "max": 9480.0, "mean": 9480.0, "n": 7 }, "utilization_pct": { "min": 6.0, "median": 42.0, "max": 43.0, "mean": 26.9, "n": 7 }, "temperature_c": { "min": 52.0, "median": 56.0, "max": 56.0, "mean": 55.0, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 60.0, "median": 60.0, "max": 60.0, "mean": 60.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1890.0, "median": 1980.0, "max": 1980.0, "mean": 1967.1, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.2857, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.2857, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.68, "median": 114.39, "max": 218.98, "mean": 117.24, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8970.0, "median": 8970.0, "max": 8970.0, "mean": 8970.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 57.0, "mean": 33.4, "n": 10 }, "temperature_c": { "min": 48.0, "median": 53.0, "max": 54.0, "mean": 51.8, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1965.0, "max": 1965.0, "mean": 1597.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 103.68, "median": 213.09, "max": 218.98, "mean": 182.14, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8970.0, "median": 8970.0, "max": 8970.0, "mean": 8970.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 55.0, "max": 57.0, "mean": 55.7, "n": 6 }, "temperature_c": { "min": 53.0, "median": 53.5, "max": 54.0, "mean": 53.5, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1198.57, "watts_per_tok_s": 1.1986, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 56.0, "temp_min_c": 52.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 60.0, "fan_max_pct": 60.0, "clock_floor_mhz": 1890.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 149.63, "power_pct_of_limit": 49.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.2857, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.2857, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 56.0, "temp_min_c": 52.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 60.0, "fan_max_pct": 60.0, "clock_floor_mhz": 1890.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 149.63, "power_pct_of_limit": 49.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.2857, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.2857, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 54.0, "temp_min_c": 53.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 182.14, "power_pct_of_limit": 60.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9480.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8970.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 124.623, "median": 124.702, "max": 124.841, "mean": 124.722, "n": 3 }, "ttft_ms": { "min": 562.11, "median": 575.17, "max": 611.99, "mean": 583.09, "n": 3 }, "power_mean_w": { "min": 149.63, "median": 169.38, "max": 176.19, "mean": 165.07, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1198.57, "median": 1358.27, "max": 1413.79, "mean": 1323.54, "n": 3 } }, "finished_utc": "2026-09-17T00:45:12Z" }, { "num_ctx": 65536, "started_utc": "2026-09-17T00:45:12Z", "loads": true, "size_total": 18204852222, "size_vram": 18204852222, "context_length_loaded": 65536, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 196.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 230.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9676.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9200.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9480.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8970.0 }, "runner_peak_rss_gib": 2.77, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 20404, "rss_kb": 2899432, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 5125 21791 315 20967 41915\nSwap: 24575 0 24575", "kv_footprint_bytes": 965990154, "kv_footprint_gib": 0.9, "runs": [ { "ttft_ms": 577.04, "first_token_was_thinking": false, "wall_s": 2.6357, "eval_count": 256, "eval_duration_ns": 2055681999, "decode_tok_s": 124.533, "prompt_eval_count": 536, "prompt_eval_duration_ns": 113423000, "prefill_tok_s": 4725.673, "load_duration_ns": 459050785, "total_duration_ns": 2632396712, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.866, "decode_tok_s_wall_gross": 97.127, "num_ctx_option": 65536, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 103.87, "median": 103.88, "max": 104.03, "mean": 103.93, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9676.0, "median": 9676.0, "max": 9676.0, "mean": 9676.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 168.47, "max_w": 203.57, "median_w": 185.96, "mean_w_whole_window": 149.13, "max_w_whole_window": 203.57, "limit_w": 300.0, "over_idle_w": 64.54, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 103.99, "median": 138.78, "max": 203.57, "mean": 149.13, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9676.0, "median": 9676.0, "max": 9676.0, "mean": 9676.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 24.0, "max": 43.0, "mean": 24.4, "n": 10 }, "temperature_c": { "min": 52.0, "median": 56.5, "max": 58.0, "mean": 55.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 60.0, "median": 60.0, "max": 60.0, "mean": 60.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1980.0, "mean": 1903.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 111.08, "median": 185.96, "max": 203.57, "mean": 168.47, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9676.0, "median": 9676.0, "max": 9676.0, "mean": 9676.0, "n": 7 }, "utilization_pct": { "min": 24.0, "median": 43.0, "max": 43.0, "mean": 34.9, "n": 7 }, "temperature_c": { "min": 54.0, "median": 57.0, "max": 58.0, "mean": 56.9, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 60.0, "median": 60.0, "max": 60.0, "mean": 60.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1980.0, "mean": 1947.9, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.31, "median": 148.7, "max": 219.7, "mean": 153.27, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9200.0, "median": 9200.0, "max": 9200.0, "mean": 9200.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 54.5, "max": 57.0, "mean": 33.7, "n": 10 }, "temperature_c": { "min": 51.0, "median": 54.5, "max": 56.0, "mean": 53.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1915.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 140.79, "median": 211.34, "max": 219.7, "mean": 193.19, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9200.0, "median": 9200.0, "max": 9200.0, "mean": 9200.0, "n": 6 }, "utilization_pct": { "min": 54.0, "median": 57.0, "max": 57.0, "mean": 56.2, "n": 6 }, "temperature_c": { "min": 54.0, "median": 55.0, "max": 56.0, "mean": 55.0, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1352.82, "watts_per_tok_s": 1.3528, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 58.0, "temp_min_c": 54.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 60.0, "fan_max_pct": 60.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 168.47, "power_pct_of_limit": 56.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 56.2% of the cap and the card at 58 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 58.0, "temp_min_c": 54.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 60.0, "fan_max_pct": 60.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 168.47, "power_pct_of_limit": 56.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 56.2% of the cap and the card at 58 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 56.0, "temp_min_c": 54.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 193.19, "power_pct_of_limit": 64.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9676.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9200.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 613.4, "first_token_was_thinking": false, "wall_s": 2.6729, "eval_count": 256, "eval_duration_ns": 2056472000, "decode_tok_s": 124.485, "prompt_eval_count": 536, "prompt_eval_duration_ns": 143305000, "prefill_tok_s": 3740.274, "load_duration_ns": 465571347, "total_duration_ns": 2669456249, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.819, "decode_tok_s_wall_gross": 95.778, "num_ctx_option": 65536, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.35, "median": 16.73, "max": 17.05, "mean": 16.71, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9455.0, "median": 9455.0, "max": 9455.0, "mean": 9455.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 3, "mean_w": 201.74, "max_w": 202.89, "median_w": 202.75, "mean_w_whole_window": 108.27, "max_w_whole_window": 202.89, "limit_w": 300.0, "over_idle_w": 185.03, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 16.51, "median": 90.43, "max": 202.89, "mean": 108.27, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9455.0, "median": 9455.0, "max": 9455.0, "mean": 9455.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 43.0, "mean": 12.7, "n": 10 }, "temperature_c": { "min": 47.0, "median": 55.5, "max": 57.0, "mean": 53.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 61.0, "median": 61.0, "max": 61.0, "mean": 61.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1980.0, "max": 1980.0, "mean": 1599.0, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 199.59, "median": 202.75, "max": 202.89, "mean": 201.74, "n": 3 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9455.0, "median": 9455.0, "max": 9455.0, "mean": 9455.0, "n": 3 }, "utilization_pct": { "min": 41.0, "median": 43.0, "max": 43.0, "mean": 42.3, "n": 3 }, "temperature_c": { "min": 57.0, "median": 57.0, "max": 57.0, "mean": 57.0, "n": 3 }, "memory_temperature_c": null, "fan_pct": { "min": 61.0, "median": 61.0, "max": 61.0, "mean": 61.0, "n": 3 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 3 }, "clocks_sm_mhz": { "min": 1980.0, "median": 1980.0, "max": 1980.0, "mean": 1980.0, "n": 3 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 3 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 3 }, "busy_samples": 3 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.61, "median": 93.67, "max": 218.94, "mean": 114.59, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9200.0, "median": 9200.0, "max": 9200.0, "mean": 9200.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 56.0, "mean": 16.8, "n": 10 }, "temperature_c": { "min": 48.0, "median": 53.5, "max": 55.0, "mean": 52.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 210.0, "median": 1965.0, "max": 1965.0, "mean": 1597.5, "n": 10 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 7481.8, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 218.34, "median": 218.53, "max": 218.94, "mean": 218.6, "n": 3 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9200.0, "median": 9200.0, "max": 9200.0, "mean": 9200.0, "n": 3 }, "utilization_pct": { "min": 56.0, "median": 56.0, "max": 56.0, "mean": 56.0, "n": 3 }, "temperature_c": { "min": 54.0, "median": 55.0, "max": 55.0, "mean": 54.7, "n": 3 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 3 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 3 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 3 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 3 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 3 }, "busy_samples": 3 } } }, "energy_j_per_1k_tokens": 1620.6, "watts_per_tok_s": 1.6206, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 3, "temp_max_c": 57.0, "temp_min_c": 57.0, "temp_rise_c": 0.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 61.0, "fan_max_pct": 61.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 201.74, "power_pct_of_limit": 67.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 3, "temp_max_c": 57.0, "temp_min_c": 57.0, "temp_rise_c": 0.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 61.0, "fan_max_pct": 61.0, "clock_floor_mhz": 1980.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1980.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 201.74, "power_pct_of_limit": 67.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 3, "temp_max_c": 55.0, "temp_min_c": 54.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 218.6, "power_pct_of_limit": 72.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9455.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9200.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 566.79, "first_token_was_thinking": false, "wall_s": 2.6238, "eval_count": 256, "eval_duration_ns": 2053864000, "decode_tok_s": 124.643, "prompt_eval_count": 536, "prompt_eval_duration_ns": 101878000, "prefill_tok_s": 5261.195, "load_duration_ns": 460429541, "total_duration_ns": 2620256235, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.965, "decode_tok_s_wall_gross": 97.567, "num_ctx_option": 65536, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 103.96, "median": 103.96, "max": 104.08, "mean": 104.0, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9455.0, "median": 9455.0, "max": 9455.0, "mean": 9455.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 171.18, "max_w": 203.9, "median_w": 189.23, "mean_w_whole_window": 151.22, "max_w_whole_window": 203.9, "limit_w": 300.0, "over_idle_w": 67.18, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 104.26, "median": 143.36, "max": 203.9, "mean": 151.22, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9455.0, "median": 9455.0, "max": 9455.0, "mean": 9455.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 41.0, "max": 42.0, "mean": 29.1, "n": 10 }, "temperature_c": { "min": 53.0, "median": 57.0, "max": 58.0, "mean": 56.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 61.0, "median": 61.0, "max": 61.0, "mean": 61.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1980.0, "mean": 1909.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 114.33, "median": 189.23, "max": 203.9, "mean": 171.18, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9455.0, "median": 9455.0, "max": 9455.0, "mean": 9455.0, "n": 7 }, "utilization_pct": { "min": 41.0, "median": 42.0, "max": 42.0, "mean": 41.6, "n": 7 }, "temperature_c": { "min": 55.0, "median": 58.0, "max": 58.0, "mean": 57.3, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 61.0, "median": 61.0, "max": 61.0, "mean": 61.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1965.0, "max": 1980.0, "mean": 1956.4, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.19, "median": 151.14, "max": 220.09, "mean": 155.03, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9200.0, "median": 9200.0, "max": 9200.0, "mean": 9200.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.0, "max": 57.0, "mean": 39.1, "n": 10 }, "temperature_c": { "min": 51.0, "median": 55.0, "max": 56.0, "mean": 54.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1957.5, "max": 1965.0, "mean": 1912.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 103.66, "median": 214.7, "max": 220.09, "mean": 182.81, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9200.0, "median": 9200.0, "max": 9200.0, "mean": 9200.0, "n": 7 }, "utilization_pct": { "min": 55.0, "median": 55.0, "max": 57.0, "mean": 55.9, "n": 7 }, "temperature_c": { "min": 54.0, "median": 55.0, "max": 56.0, "mean": 55.1, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1965.0, "max": 1965.0, "mean": 1960.7, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1373.36, "watts_per_tok_s": 1.3734, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 58.0, "temp_min_c": 55.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 61.0, "fan_max_pct": 61.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 171.18, "power_pct_of_limit": 57.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 57.1% of the cap and the card at 58 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 58.0, "temp_min_c": 55.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 61.0, "fan_max_pct": 61.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1980.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 171.18, "power_pct_of_limit": 57.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 57.1% of the cap and the card at 58 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 56.0, "temp_min_c": 54.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 182.81, "power_pct_of_limit": 60.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9455.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9200.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 124.485, "median": 124.533, "max": 124.643, "mean": 124.554, "n": 3 }, "ttft_ms": { "min": 566.79, "median": 577.04, "max": 613.4, "mean": 585.74, "n": 3 }, "power_mean_w": { "min": 168.47, "median": 171.18, "max": 201.74, "mean": 180.46, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1352.82, "median": 1373.36, "max": 1620.6, "mean": 1448.93, "n": 3 } }, "finished_utc": "2026-09-17T00:46:24Z" }, { "num_ctx": 98304, "started_utc": "2026-09-17T00:46:24Z", "loads": true, "size_total": 18741723134, "size_vram": 18741723134, "context_length_loaded": 98304, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 392.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 460.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9847.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9660.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9455.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9200.0 }, "runner_peak_rss_gib": 2.89, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 21181, "rss_kb": 3031208, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 4617 22298 435 21088 42422\nSwap: 24575 0 24575", "kv_footprint_bytes": 1502861066, "kv_footprint_gib": 1.4, "runs": [ { "ttft_ms": 577.73, "first_token_was_thinking": false, "wall_s": 2.6326, "eval_count": 256, "eval_duration_ns": 2051945000, "decode_tok_s": 124.76, "prompt_eval_count": 536, "prompt_eval_duration_ns": 113487000, "prefill_tok_s": 4723.008, "load_duration_ns": 459882648, "total_duration_ns": 2629225398, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.095, "decode_tok_s_wall_gross": 97.242, "num_ctx_option": 98304, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 105.64, "median": 105.67, "max": 105.68, "mean": 105.66, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 181.29, "max_w": 204.06, "median_w": 193.81, "mean_w_whole_window": 151.86, "max_w_whole_window": 204.06, "limit_w": 300.0, "over_idle_w": 75.63, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 105.66, "median": 146.08, "max": 204.06, "mean": 151.86, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 41.0, "max": 42.0, "mean": 25.0, "n": 10 }, "temperature_c": { "min": 54.0, "median": 58.0, "max": 59.0, "mean": 57.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 62.0, "median": 62.0, "max": 62.0, "mean": 62.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1911.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 139.66, "median": 193.81, "max": 204.06, "mean": 181.29, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 6 }, "utilization_pct": { "min": 41.0, "median": 42.0, "max": 42.0, "mean": 41.7, "n": 6 }, "temperature_c": { "min": 58.0, "median": 59.0, "max": 59.0, "mean": 58.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 62.0, "median": 62.0, "max": 62.0, "mean": 62.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 91.09, "median": 144.99, "max": 219.49, "mean": 153.77, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9660.0, "median": 9660.0, "max": 9660.0, "mean": 9660.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 55.5, "max": 57.0, "mean": 33.8, "n": 10 }, "temperature_c": { "min": 53.0, "median": 56.0, "max": 57.0, "mean": 55.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 136.83, "median": 214.81, "max": 219.49, "mean": 193.06, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9660.0, "median": 9660.0, "max": 9660.0, "mean": 9660.0, "n": 6 }, "utilization_pct": { "min": 55.0, "median": 56.5, "max": 57.0, "mean": 56.3, "n": 6 }, "temperature_c": { "min": 56.0, "median": 56.0, "max": 57.0, "mean": 56.2, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1453.11, "watts_per_tok_s": 1.4531, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 59.0, "temp_min_c": 58.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 62.0, "fan_max_pct": 62.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 181.29, "power_pct_of_limit": 60.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 59.0, "temp_min_c": 58.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 62.0, "fan_max_pct": 62.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 181.29, "power_pct_of_limit": 60.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 57.0, "temp_min_c": 56.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 193.06, "power_pct_of_limit": 64.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9847.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9660.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 572.44, "first_token_was_thinking": false, "wall_s": 2.6271, "eval_count": 256, "eval_duration_ns": 2051511000, "decode_tok_s": 124.786, "prompt_eval_count": 536, "prompt_eval_duration_ns": 99471000, "prefill_tok_s": 5388.505, "load_duration_ns": 468847449, "total_duration_ns": 2623771350, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 124.108, "decode_tok_s_wall_gross": 97.446, "num_ctx_option": 98304, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 106.13, "median": 106.18, "max": 106.19, "mean": 106.17, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 192.13, "max_w": 205.04, "median_w": 202.88, "mean_w_whole_window": 153.51, "max_w_whole_window": 205.04, "limit_w": 300.0, "over_idle_w": 85.96, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.08, "median": 148.19, "max": 205.04, "mean": 153.51, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 20.0, "max": 44.0, "mean": 20.8, "n": 10 }, "temperature_c": { "min": 54.0, "median": 58.5, "max": 60.0, "mean": 57.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 63.0, "median": 63.0, "max": 63.0, "mean": 63.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1965.0, "max": 1965.0, "mean": 1911.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 155.02, "median": 202.88, "max": 205.04, "mean": 192.13, "n": 5 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 5 }, "utilization_pct": { "min": 40.0, "median": 42.0, "max": 44.0, "mean": 41.6, "n": 5 }, "temperature_c": { "min": 59.0, "median": 60.0, "max": 60.0, "mean": 59.6, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 63.0, "median": 63.0, "max": 63.0, "mean": 63.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1965.0, "median": 1965.0, "max": 1965.0, "mean": 1965.0, "n": 5 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 91.44, "median": 153.07, "max": 220.2, "mean": 156.12, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9660.0, "median": 9660.0, "max": 9660.0, "mean": 9660.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 27.5, "max": 58.0, "mean": 28.5, "n": 10 }, "temperature_c": { "min": 53.0, "median": 56.0, "max": 57.0, "mean": 55.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 160.49, "median": 219.73, "max": 220.2, "mean": 207.09, "n": 5 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9660.0, "median": 9660.0, "max": 9660.0, "mean": 9660.0, "n": 5 }, "utilization_pct": { "min": 55.0, "median": 57.0, "max": 58.0, "mean": 57.0, "n": 5 }, "temperature_c": { "min": 56.0, "median": 57.0, "max": 57.0, "mean": 56.6, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 5 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 5 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1539.68, "watts_per_tok_s": 1.5397, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 60.0, "temp_min_c": 59.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 63.0, "fan_max_pct": 63.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 192.13, "power_pct_of_limit": 64.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 60.0, "temp_min_c": 59.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 63.0, "fan_max_pct": 63.0, "clock_floor_mhz": 1965.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 192.13, "power_pct_of_limit": 64.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 57.0, "temp_min_c": 56.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 207.09, "power_pct_of_limit": 69.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9847.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9660.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 562.63, "first_token_was_thinking": false, "wall_s": 2.6191, "eval_count": 256, "eval_duration_ns": 2053444000, "decode_tok_s": 124.669, "prompt_eval_count": 536, "prompt_eval_duration_ns": 99030000, "prefill_tok_s": 5412.501, "load_duration_ns": 459269439, "total_duration_ns": 2615844617, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 123.998, "decode_tok_s_wall_gross": 97.743, "num_ctx_option": 98304, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 106.08, "median": 106.09, "max": 106.09, "mean": 106.09, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 172.18, "max_w": 204.84, "median_w": 188.17, "mean_w_whole_window": 152.43, "max_w_whole_window": 204.84, "limit_w": 300.0, "over_idle_w": 66.09, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.06, "median": 145.93, "max": 204.84, "mean": 152.43, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 41.0, "max": 43.0, "mean": 29.3, "n": 10 }, "temperature_c": { "min": 55.0, "median": 58.5, "max": 60.0, "mean": 57.9, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 63.0, "median": 63.0, "max": 64.0, "mean": 63.1, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1957.5, "max": 1965.0, "mean": 1900.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 115.58, "median": 188.17, "max": 204.84, "mean": 172.18, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9847.0, "median": 9847.0, "max": 9847.0, "mean": 9847.0, "n": 7 }, "utilization_pct": { "min": 39.0, "median": 43.0, "max": 43.0, "mean": 41.9, "n": 7 }, "temperature_c": { "min": 57.0, "median": 60.0, "max": 60.0, "mean": 59.1, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 63.0, "median": 63.0, "max": 64.0, "mean": 63.1, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1830.0, "median": 1965.0, "max": 1965.0, "mean": 1943.6, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 91.65, "median": 151.51, "max": 220.65, "mean": 155.91, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9660.0, "median": 9660.0, "max": 9660.0, "mean": 9660.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 50.0, "max": 56.0, "mean": 36.7, "n": 10 }, "temperature_c": { "min": 53.0, "median": 57.0, "max": 57.0, "mean": 55.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 104.66, "median": 215.25, "max": 220.65, "mean": 183.44, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9660.0, "median": 9660.0, "max": 9660.0, "mean": 9660.0, "n": 7 }, "utilization_pct": { "min": 45.0, "median": 55.0, "max": 56.0, "mean": 52.4, "n": 7 }, "temperature_c": { "min": 56.0, "median": 57.0, "max": 57.0, "mean": 56.9, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1381.1, "watts_per_tok_s": 1.3811, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 60.0, "temp_min_c": 57.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 63.1, "fan_max_pct": 64.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 172.18, "power_pct_of_limit": 57.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 57.4% of the cap and the card at 60 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 60.0, "temp_min_c": 57.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 63.1, "fan_max_pct": 64.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1965.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 300.0, "power_mean_w": 172.18, "power_pct_of_limit": 57.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "NEITHER cap: the SM clock varied with draw at 57.4% of the cap and the card at 60 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 57.0, "temp_min_c": 56.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 183.44, "power_pct_of_limit": 61.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9847.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9660.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 124.669, "median": 124.76, "max": 124.786, "mean": 124.738, "n": 3 }, "ttft_ms": { "min": 562.63, "median": 572.44, "max": 577.73, "mean": 570.93, "n": 3 }, "power_mean_w": { "min": 172.18, "median": 181.29, "max": 192.13, "mean": 181.87, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1381.1, "median": 1453.11, "max": 1539.68, "mean": 1457.96, "n": 3 } }, "finished_utc": "2026-09-17T00:47:26Z" }, { "num_ctx": 131072, "started_utc": "2026-09-17T00:47:26Z", "loads": true, "size_total": 18412428326, "size_vram": 18412428326, "context_length_loaded": 131072, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": -22.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 46.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9825.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9706.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9847.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9660.0 }, "runner_peak_rss_gib": 2.64, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 22052, "rss_kb": 2768240, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 4379 22537 171 20823 42661\nSwap: 24575 0 24575", "kv_footprint_bytes": 1173566258, "kv_footprint_gib": 1.09, "runs": [ { "ttft_ms": 578.5, "first_token_was_thinking": false, "wall_s": 2.8174, "eval_count": 256, "eval_duration_ns": 2235850000, "decode_tok_s": 114.498, "prompt_eval_count": 536, "prompt_eval_duration_ns": 114532000, "prefill_tok_s": 4679.915, "load_duration_ns": 459427513, "total_duration_ns": 2814037880, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 113.893, "decode_tok_s_wall_gross": 90.862, "num_ctx_option": 131072, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 106.45, "median": 106.47, "max": 106.5, "mean": 106.47, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 151.92, "max_w": 160.83, "median_w": 160.43, "mean_w_whole_window": 134.5, "max_w_whole_window": 160.83, "limit_w": 300.0, "over_idle_w": 45.45, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.49, "median": 134.59, "max": 160.83, "mean": 134.5, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 38.0, "max": 41.0, "mean": 23.3, "n": 10 }, "temperature_c": { "min": 55.0, "median": 58.0, "max": 58.0, "mean": 57.1, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 64.0, "mean": 64.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1800.0, "max": 1800.0, "mean": 1800.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 131.8, "median": 160.43, "max": 160.83, "mean": 151.92, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 6 }, "utilization_pct": { "min": 38.0, "median": 38.5, "max": 41.0, "mean": 38.8, "n": 6 }, "temperature_c": { "min": 58.0, "median": 58.0, "max": 58.0, "mean": 58.0, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 64.0, "mean": 64.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1800.0, "max": 1800.0, "mean": 1800.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 91.47, "median": 141.94, "max": 211.99, "mean": 150.08, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9706.0, "median": 9706.0, "max": 9706.0, "mean": 9706.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 25.0, "max": 54.0, "mean": 26.0, "n": 10 }, "temperature_c": { "min": 54.0, "median": 57.0, "max": 58.0, "mean": 56.0, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 149.62, "median": 211.14, "max": 211.99, "mean": 197.31, "n": 5 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9706.0, "median": 9706.0, "max": 9706.0, "mean": 9706.0, "n": 5 }, "utilization_pct": { "min": 50.0, "median": 53.0, "max": 54.0, "mean": 52.0, "n": 5 }, "temperature_c": { "min": 57.0, "median": 57.0, "max": 58.0, "mean": 57.2, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 5 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 5 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1326.84, "watts_per_tok_s": 1.3268, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 58.0, "temp_min_c": 58.0, "temp_rise_c": 0.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 64.0, "fan_max_pct": 64.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1800.0, "clock_median_mhz": 1800.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 151.92, "power_pct_of_limit": 50.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 58.0, "temp_min_c": 58.0, "temp_rise_c": 0.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 64.0, "fan_max_pct": 64.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1800.0, "clock_median_mhz": 1800.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 151.92, "power_pct_of_limit": 50.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 58.0, "temp_min_c": 57.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 197.31, "power_pct_of_limit": 65.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9825.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9706.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 13.0, "median": 13.0, "max": 13.0, "mean": 13.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 571.4, "first_token_was_thinking": false, "wall_s": 2.7911, "eval_count": 256, "eval_duration_ns": 2216636000, "decode_tok_s": 115.49, "prompt_eval_count": 536, "prompt_eval_duration_ns": 101885000, "prefill_tok_s": 5260.833, "load_duration_ns": 465248514, "total_duration_ns": 2787818611, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 114.881, "decode_tok_s_wall_gross": 91.72, "num_ctx_option": 131072, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 106.52, "median": 106.57, "max": 106.66, "mean": 106.58, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 155.19, "max_w": 167.75, "median_w": 164.01, "mean_w_whole_window": 136.44, "max_w_whole_window": 167.75, "limit_w": 300.0, "over_idle_w": 48.61, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.53, "median": 134.71, "max": 167.75, "mean": 136.44, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 38.0, "max": 40.0, "mean": 23.4, "n": 10 }, "temperature_c": { "min": 56.0, "median": 58.0, "max": 59.0, "mean": 57.8, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 64.0, "mean": 64.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1830.0, "max": 1875.0, "mean": 1828.5, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 131.42, "median": 164.01, "max": 167.75, "mean": 155.19, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 6 }, "utilization_pct": { "min": 38.0, "median": 39.0, "max": 40.0, "mean": 39.0, "n": 6 }, "temperature_c": { "min": 58.0, "median": 59.0, "max": 59.0, "mean": 58.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 64.0, "mean": 64.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1815.0, "median": 1845.0, "max": 1875.0, "mean": 1847.5, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 92.09, "median": 150.76, "max": 213.46, "mean": 152.6, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9706.0, "median": 9706.0, "max": 9706.0, "mean": 9706.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 51.0, "max": 53.0, "mean": 31.2, "n": 10 }, "temperature_c": { "min": 54.0, "median": 57.0, "max": 58.0, "mean": 56.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 143.85, "median": 207.55, "max": 213.46, "mean": 190.55, "n": 6 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9706.0, "median": 9706.0, "max": 9706.0, "mean": 9706.0, "n": 6 }, "utilization_pct": { "min": 51.0, "median": 52.0, "max": 53.0, "mean": 52.0, "n": 6 }, "temperature_c": { "min": 57.0, "median": 58.0, "max": 58.0, "mean": 57.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 6 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 6 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1343.75, "watts_per_tok_s": 1.3438, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 59.0, "temp_min_c": 58.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 64.0, "fan_max_pct": 64.0, "clock_floor_mhz": 1815.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1845.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 155.19, "power_pct_of_limit": 51.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 59.0, "temp_min_c": 58.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 64.0, "fan_max_pct": 64.0, "clock_floor_mhz": 1815.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1845.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 155.19, "power_pct_of_limit": 51.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 58.0, "temp_min_c": 57.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 190.55, "power_pct_of_limit": 63.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9825.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9706.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 14.0, "median": 14.0, "max": 14.0, "mean": 14.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 566.57, "first_token_was_thinking": false, "wall_s": 2.7885, "eval_count": 256, "eval_duration_ns": 2219028000, "decode_tok_s": 115.366, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102113000, "prefill_tok_s": 5249.087, "load_duration_ns": 460132716, "total_duration_ns": 2785352103, "streamed_chunks": 256, "response_chars": 1042, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 114.764, "decode_tok_s_wall_gross": 91.805, "num_ctx_option": 131072, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 106.69, "median": 106.71, "max": 106.71, "mean": 106.7, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 148.88, "max_w": 167.71, "median_w": 161.99, "mean_w_whole_window": 136.23, "max_w_whole_window": 167.71, "limit_w": 300.0, "over_idle_w": 42.18, "samples": 10, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.68, "median": 134.88, "max": 167.71, "mean": 136.23, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 39.0, "max": 41.0, "mean": 27.8, "n": 10 }, "temperature_c": { "min": 56.0, "median": 58.0, "max": 59.0, "mean": 57.8, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 64.0, "mean": 64.0, "n": 10 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1807.5, "max": 1860.0, "mean": 1821.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 114.03, "median": 161.99, "max": 167.71, "mean": 148.88, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9825.0, "median": 9825.0, "max": 9825.0, "mean": 9825.0, "n": 7 }, "utilization_pct": { "min": 38.0, "median": 40.0, "max": 41.0, "mean": 39.7, "n": 7 }, "temperature_c": { "min": 58.0, "median": 59.0, "max": 59.0, "mean": 58.6, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 64.0, "mean": 64.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1830.0, "max": 1860.0, "mean": 1830.0, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 92.06, "median": 151.21, "max": 213.26, "mean": 152.74, "n": 10 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9706.0, "median": 9706.0, "max": 9706.0, "mean": 9706.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 52.0, "max": 53.0, "mean": 36.7, "n": 10 }, "temperature_c": { "min": 55.0, "median": 57.5, "max": 58.0, "mean": 56.9, "n": 10 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 10 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1905.0, "n": 10 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.14, "median": 203.7, "max": 213.26, "mean": 178.7, "n": 7 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9706.0, "median": 9706.0, "max": 9706.0, "mean": 9706.0, "n": 7 }, "utilization_pct": { "min": 51.0, "median": 53.0, "max": 53.0, "mean": 52.4, "n": 7 }, "temperature_c": { "min": 57.0, "median": 58.0, "max": 58.0, "mean": 57.7, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 7 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 7 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1290.5, "watts_per_tok_s": 1.2905, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 59.0, "temp_min_c": 58.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 64.0, "fan_max_pct": 64.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1830.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 148.88, "power_pct_of_limit": 49.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 59.0, "temp_min_c": 58.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 64.0, "fan_max_pct": 64.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1830.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 148.88, "power_pct_of_limit": 49.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 58.0, "temp_min_c": 57.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 4.0, "pcie_width_max": 4.0, "pcie_width_median": 4.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1950.0, "clock_median_mhz": 1950.0, "mem_clock_floor_mhz": 9251.0, "mem_clock_max_mhz": 9251.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 300.0, "power_mean_w": 178.7, "power_pct_of_limit": 59.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9825.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9706.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 10 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 10 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 2, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 2 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 114.498, "median": 115.366, "max": 115.49, "mean": 115.118, "n": 3 }, "ttft_ms": { "min": 566.57, "median": 571.4, "max": 578.5, "mean": 572.16, "n": 3 }, "power_mean_w": { "min": 148.88, "median": 151.92, "max": 155.19, "mean": 152.0, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1290.5, "median": 1326.84, "max": 1343.75, "mean": 1320.36, "n": 3 } }, "finished_utc": "2026-09-17T00:48:29Z" } ], "errors": [], "context_length_max": 262144, "quantization": "Q4_K_M", "ups_after_arm": { "ups": "pr1500@localhost", "read_utc": "2026-09-17T00:48:30Z", "available": true, "raw": "battery.charge: 100\nbattery.charge.low: 10\nbattery.charge.warning: 20\nbattery.mfr.date: CPS\nbattery.runtime: 3675\nbattery.runtime.low: 300\nbattery.status: 100%\nbattery.type: PbAcid\nbattery.voltage: 27.5\nbattery.voltage.nominal: 24\ndevice.mfr: CPS\ndevice.model: CP1500PFCLCDa\ndevice.serial: CXXRS7000436\ndevice.type: ups\ndriver.debug: 0\ndriver.flag.allow_killpower: 0\ndriver.name: usbhid-ups\ndriver.parameter.interrupt_pipe_no_events_tolerance: -1\ndriver.parameter.pollfreq: 12\ndriver.parameter.pollinterval: 2\ndriver.parameter.port: auto\ndriver.parameter.synchronous: auto\ndriver.state: quiet\ndriver.version: 2.8.4\ndriver.version.data: CyberPower HID 0.84\ndriver.version.internal: 0.67\ndriver.version.usb: libusb-1.0.29 (API: 0x0100010B)\ninput.sensitivity: normal\ninput.transfer.high: 139\ninput.transfer.low: 100\ninput.voltage: 121.0\ninput.voltage.nominal: 120\noutput.voltage: 121.0\nups.beeper.status: enabled\nups.delay.shutdown: 60\nups.delay.start: 120\nups.firmware: CR01802H1\nups.load: 12\nups.mfr: CPS\nups.model: CP1500PFCLCDa\nups.power: 124\nups.power.nominal: 1500\nups.productid: 0601\nups.realpower: 124\nups.realpower.nominal: 1000\nups.serial: CXXRS7000436\nups.status: OL\nups.test.result: No test initiated\nups.timer.shutdown: -60\nups.timer.start: -60\nups.vendorid: 0764", "battery.charge": 100.0, "input.voltage": 121.0, "output.voltage": 121.0, "ups.load": 12.0, "ups.realpower": 124.0, "ups.realpower.nominal": 1000.0, "ups.status": "OL", "load_pct": 12.0, "stop_threshold_pct": 80.0, "over_stop_threshold": false }, "largest_context_that_fits": 131072, "levels_requested": [ 4096, 8192, 16384, 32768, 49152, 65536, 98304, 131072 ], "levels_not_reached": [], "ladder_topped_out": "every rung tested fits, up to and including 131,072 -- the ladder ran out of rungs before the cards ran out of memory, so the figure reported is the largest window TESTED, not a measured ceiling", "headline": "the biggest context this arm holds for gemma4:26b is 131,072 tokens", "finished_utc": "2026-09-17T00:48:30Z" }