{ "model_id": "m4", "model_tag": "mistral-small3.2:24b", "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": "2026-09-17T01:00:06Z", "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": "2026-09-17T01:00:06Z", "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": "2026-09-17T00:53:55Z", "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": "2026-09-17T01:00:06Z", "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: 126\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": "2026-09-17T01:00:06Z", "loads": true, "size_total": 14947555736, "size_vram": 14947555736, "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": 8297.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7329.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8329.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7346.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 32.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 17.0 }, "runner_peak_rss_gib": 1.12, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 109235, "rss_kb": 1179196, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 4031 1250 65 42371 43009\nSwap: 24575 0 24575", "kv_footprint_bytes": 0, "kv_footprint_gib": 0.0, "runs": [ { "ttft_ms": 301.74, "first_token_was_thinking": false, "wall_s": 6.0253, "eval_count": 256, "eval_duration_ns": 5721851000, "decode_tok_s": 44.741, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 73914000, "prefill_tok_s": 13840.409, "load_duration_ns": 223783996, "total_duration_ns": 6023317276, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.553, "decode_tok_s_wall_gross": 42.487, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.95, "median": 17.02, "max": 17.62, "mean": 17.2, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 265.8, "max_w": 281.69, "median_w": 280.27, "mean_w_whole_window": 231.88, "max_w_whole_window": 281.69, "limit_w": 300.0, "over_idle_w": 248.6, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.09, "median": 280.07, "max": 281.69, "mean": 231.88, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 54.0, "mean": 43.0, "n": 21 }, "temperature_c": { "min": 46.0, "median": 60.0, "max": 62.0, "mean": 58.5, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 59.0, "max": 64.0, "mean": 60.1, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1980.0, "mean": 1841.4, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.9524, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 140.22, "median": 280.27, "max": 281.69, "mean": 265.8, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 50.0, "max": 54.0, "mean": 50.2, "n": 18 }, "temperature_c": { "min": 57.0, "median": 60.0, "max": 62.0, "mean": 60.1, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 58.0, "median": 60.0, "max": 64.0, "mean": 60.5, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1950.0, "max": 1980.0, "mean": 1956.7, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 10.93, "median": 255.59, "max": 258.72, "mean": 210.36, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7346.0, "median": 7346.0, "max": 7346.0, "mean": 7346.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 47.0, "max": 54.0, "mean": 42.0, "n": 21 }, "temperature_c": { "min": 42.0, "median": 51.0, "max": 53.0, "mean": 50.2, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1965.0, "max": 1965.0, "mean": 1865.0, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.0952, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0952, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 79.65, "median": 257.26, "max": 258.72, "mean": 241.42, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7346.0, "median": 7346.0, "max": 7346.0, "mean": 7346.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 49.0, "max": 54.0, "mean": 48.9, "n": 18 }, "temperature_c": { "min": 49.0, "median": 51.5, "max": 53.0, "mean": 51.2, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1965.0, "max": 1965.0, "mean": 1956.7, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 0.1111, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.1111, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 5940.89, "power_mean_w_all_cards": 507.22, "energy_j_per_1k_tokens_all_cards": 11336.86, "watts_per_tok_s": 5.9409, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 62.0, "temp_min_c": 57.0, "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": 60.5, "fan_max_pct": 64.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1980.0, "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": 265.8, "power_pct_of_limit": 88.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "temp_max_c": 62.0, "temp_min_c": 57.0, "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": 60.5, "fan_max_pct": 64.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1980.0, "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": 265.8, "power_pct_of_limit": 88.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "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": 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": 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": 241.42, "power_pct_of_limit": 80.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.1111, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1111, "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": 8329.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7346.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 285.93, "first_token_was_thinking": false, "wall_s": 6.012, "eval_count": 256, "eval_duration_ns": 5724382000, "decode_tok_s": 44.721, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 57049000, "prefill_tok_s": 17931.953, "load_duration_ns": 224727793, "total_duration_ns": 6010023778, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.533, "decode_tok_s_wall_gross": 42.581, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.74, "median": 16.78, "max": 17.09, "mean": 16.87, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 264.2, "max_w": 281.71, "median_w": 280.62, "mean_w_whole_window": 230.9, "max_w_whole_window": 281.71, "limit_w": 300.0, "over_idle_w": 247.33, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.12, "median": 280.29, "max": 281.71, "mean": 230.9, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 50.0, "max": 54.0, "mean": 42.9, "n": 21 }, "temperature_c": { "min": 46.0, "median": 61.0, "max": 62.0, "mean": 59.0, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 65.0, "mean": 64.5, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1965.0, "max": 1980.0, "mean": 1858.6, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.9524, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0952, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 85.08, "median": 280.62, "max": 281.71, "mean": 264.2, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 18 }, "utilization_pct": { "min": 47.0, "median": 50.0, "max": 54.0, "mean": 50.1, "n": 18 }, "temperature_c": { "min": 59.0, "median": 61.0, "max": 62.0, "mean": 60.7, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 64.0, "median": 64.0, "max": 65.0, "mean": 64.4, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1965.0, "max": 1980.0, "mean": 1960.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.1111, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.2, "median": 256.55, "max": 258.16, "mean": 211.61, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7346.0, "median": 7346.0, "max": 7346.0, "mean": 7346.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 47.0, "max": 54.0, "mean": 42.0, "n": 21 }, "temperature_c": { "min": 43.0, "median": 52.0, "max": 53.0, "mean": 50.9, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1950.0, "mean": 1847.1, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.0952, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 81.56, "median": 256.8, "max": 258.16, "mean": 242.8, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7346.0, "median": 7346.0, "max": 7346.0, "mean": 7346.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 50.0, "max": 54.0, "mean": 48.9, "n": 18 }, "temperature_c": { "min": 49.0, "median": 52.0, "max": 53.0, "mean": 51.9, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1935.0, "mean": 1935.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 0.1111, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 5907.74, "power_mean_w_all_cards": 507.0, "energy_j_per_1k_tokens_all_cards": 11336.96, "watts_per_tok_s": 5.9077, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 62.0, "temp_min_c": 59.0, "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": 64.4, "fan_max_pct": 65.0, "clock_floor_mhz": 1950.0, "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": false, "power_limit_w": 300.0, "power_mean_w": 264.2, "power_pct_of_limit": 88.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1111, "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": 18, "temp_max_c": 62.0, "temp_min_c": 59.0, "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": 64.4, "fan_max_pct": 65.0, "clock_floor_mhz": 1950.0, "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": false, "power_limit_w": 300.0, "power_mean_w": 264.2, "power_pct_of_limit": 88.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1111, "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": 18, "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": 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": 242.8, "power_pct_of_limit": 80.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.1111, "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": 8329.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7346.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 281.87, "first_token_was_thinking": false, "wall_s": 6.01, "eval_count": 256, "eval_duration_ns": 5726321000, "decode_tok_s": 44.706, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25111000, "prefill_tok_s": 40739.118, "load_duration_ns": 253009037, "total_duration_ns": 6008036629, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.517, "decode_tok_s_wall_gross": 42.596, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 105.1, "median": 105.12, "max": 105.13, "mean": 105.12, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 270.41, "max_w": 283.58, "median_w": 281.82, "mean_w_whole_window": 247.85, "max_w_whole_window": 283.58, "limit_w": 300.0, "over_idle_w": 165.29, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 105.16, "median": 281.77, "max": 283.58, "mean": 247.85, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 47.0, "max": 54.0, "mean": 41.2, "n": 21 }, "temperature_c": { "min": 53.0, "median": 62.0, "max": 64.0, "mean": 61.2, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 67.0, "median": 67.0, "max": 68.0, "mean": 67.0, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1965.0, "mean": 1942.1, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 147.58, "median": 281.82, "max": 283.58, "mean": 270.41, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8329.0, "median": 8329.0, "max": 8329.0, "mean": 8329.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 47.0, "max": 54.0, "mean": 48.1, "n": 18 }, "temperature_c": { "min": 60.0, "median": 62.0, "max": 64.0, "mean": 62.3, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 67.0, "median": 67.0, "max": 68.0, "mean": 67.1, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1965.0, "mean": 1956.7, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.69, "median": 257.22, "max": 258.69, "mean": 226.05, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7346.0, "median": 7346.0, "max": 7346.0, "mean": 7346.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 53.0, "mean": 42.3, "n": 21 }, "temperature_c": { "min": 47.0, "median": 54.0, "max": 55.0, "mean": 53.0, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1950.0, "mean": 1927.1, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.0476, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0476, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 134.64, "median": 257.5, "max": 258.69, "mean": 247.51, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7346.0, "median": 7346.0, "max": 7346.0, "mean": 7346.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 50.0, "max": 53.0, "mean": 49.4, "n": 18 }, "temperature_c": { "min": 52.0, "median": 54.0, "max": 55.0, "mean": 53.8, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1950.0, "mean": 1940.8, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 0.0556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0556, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6048.65, "power_mean_w_all_cards": 517.92, "energy_j_per_1k_tokens_all_cards": 11585.06, "watts_per_tok_s": 6.0486, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 64.0, "temp_min_c": 60.0, "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": 67.1, "fan_max_pct": 68.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1965.0, "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": 270.41, "power_pct_of_limit": 90.1, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.1% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 64.0, "temp_min_c": 60.0, "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": 67.1, "fan_max_pct": 68.0, "clock_floor_mhz": 1950.0, "clock_max_mhz": 1965.0, "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": 270.41, "power_pct_of_limit": 90.1, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.1% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "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": 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": 247.51, "power_pct_of_limit": 82.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0556, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0556, "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": 8329.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7346.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 44.706, "median": 44.721, "max": 44.741, "mean": 44.723, "n": 3 }, "ttft_ms": { "min": 281.87, "median": 285.93, "max": 301.74, "mean": 289.85, "n": 3 }, "power_mean_w": { "min": 264.2, "median": 265.8, "max": 270.41, "mean": 266.8, "n": 3 }, "power_mean_w_all_cards": { "min": 507.0, "median": 507.22, "max": 517.92, "mean": 510.71, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 11336.86, "median": 11336.96, "max": 11585.06, "mean": 11419.63, "n": 3 }, "energy_j_per_1k_tokens": { "min": 5907.74, "median": 5940.89, "max": 6048.65, "mean": 5965.76, "n": 3 } }, "finished_utc": "2026-09-17T01:01:52Z" }, { "num_ctx": 8192, "started_utc": "2026-09-17T01:01:52Z", "loads": true, "size_total": 15337626008, "size_vram": 15337626008, "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": 194.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 178.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8523.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7524.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8329.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7346.0 }, "runner_peak_rss_gib": 0.92, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 110658, "rss_kb": 966164, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 2813 2451 81 42403 44226\nSwap: 24575 0 24575", "kv_footprint_bytes": 390070272, "kv_footprint_gib": 0.36, "runs": [ { "ttft_ms": 308.47, "first_token_was_thinking": false, "wall_s": 6.0416, "eval_count": 256, "eval_duration_ns": 5731241000, "decode_tok_s": 44.667, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 64221000, "prefill_tok_s": 15929.369, "load_duration_ns": 241107507, "total_duration_ns": 6039578139, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.478, "decode_tok_s_wall_gross": 42.373, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 17.0, "median": 17.37, "max": 17.67, "mean": 17.41, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 267.98, "max_w": 283.67, "median_w": 282.86, "mean_w_whole_window": 233.85, "max_w_whole_window": 283.67, "limit_w": 300.0, "over_idle_w": 250.57, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.0, "median": 282.78, "max": 283.67, "mean": 233.85, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 51.0, "max": 54.0, "mean": 43.6, "n": 21 }, "temperature_c": { "min": 49.0, "median": 62.0, "max": 64.0, "mean": 60.8, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 66.0, "median": 66.0, "max": 67.0, "mean": 66.5, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1965.0, "mean": 1860.7, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0952, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 143.92, "median": 282.86, "max": 283.67, "mean": 267.98, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 18 }, "utilization_pct": { "min": 47.0, "median": 51.0, "max": 54.0, "mean": 50.8, "n": 18 }, "temperature_c": { "min": 59.0, "median": 62.5, "max": 64.0, "mean": 62.3, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 66.0, "median": 66.0, "max": 67.0, "mean": 66.4, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1957.5, "max": 1965.0, "mean": 1955.8, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.1111, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.65, "median": 257.57, "max": 258.9, "mean": 211.22, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7524.0, "median": 7524.0, "max": 7524.0, "mean": 7524.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 55.0, "mean": 43.1, "n": 21 }, "temperature_c": { "min": 44.0, "median": 53.0, "max": 55.0, "mean": 52.4, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1950.0, "mean": 1842.9, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.0476, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0476, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 72.71, "median": 257.84, "max": 258.9, "mean": 242.08, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7524.0, "median": 7524.0, "max": 7524.0, "mean": 7524.0, "n": 18 }, "utilization_pct": { "min": 46.0, "median": 51.0, "max": 55.0, "mean": 50.3, "n": 18 }, "temperature_c": { "min": 51.0, "median": 54.0, "max": 55.0, "mean": 53.5, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1950.0, "mean": 1936.7, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 0.0556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0556, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 5999.45, "power_mean_w_all_cards": 510.06, "energy_j_per_1k_tokens_all_cards": 11419.05, "watts_per_tok_s": 5.9995, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 64.0, "temp_min_c": 59.0, "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": 66.4, "fan_max_pct": 67.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1957.5, "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": 267.98, "power_pct_of_limit": 89.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1111, "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": 18, "temp_max_c": 64.0, "temp_min_c": 59.0, "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": 66.4, "fan_max_pct": 67.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1957.5, "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": 267.98, "power_pct_of_limit": 89.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.1111, "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": 18, "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": 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": 242.08, "power_pct_of_limit": 80.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0556, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0556, "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": 8523.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7524.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 277.22, "first_token_was_thinking": false, "wall_s": 6.0077, "eval_count": 256, "eval_duration_ns": 5728467000, "decode_tok_s": 44.689, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25455000, "prefill_tok_s": 40188.568, "load_duration_ns": 248815502, "total_duration_ns": 6005556345, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.499, "decode_tok_s_wall_gross": 42.612, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 106.33, "median": 106.34, "max": 106.4, "mean": 106.36, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 271.51, "max_w": 283.69, "median_w": 282.76, "mean_w_whole_window": 248.93, "max_w_whole_window": 283.69, "limit_w": 300.0, "over_idle_w": 165.15, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 106.43, "median": 282.63, "max": 283.69, "mean": 248.93, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 54.0, "mean": 42.8, "n": 21 }, "temperature_c": { "min": 54.0, "median": 64.0, "max": 65.0, "mean": 62.7, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 69.0, "median": 69.0, "max": 69.0, "mean": 69.0, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1755.0, "median": 1950.0, "max": 1965.0, "mean": 1932.1, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 151.37, "median": 282.76, "max": 283.69, "mean": 271.51, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 50.0, "max": 54.0, "mean": 49.9, "n": 18 }, "temperature_c": { "min": 62.0, "median": 64.0, "max": 65.0, "mean": 63.8, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 69.0, "median": 69.0, "max": 69.0, "mean": 69.0, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1950.0, "max": 1965.0, "mean": 1948.3, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.04, "median": 257.82, "max": 258.7, "mean": 226.75, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7524.0, "median": 7524.0, "max": 7524.0, "mean": 7524.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 46.0, "max": 54.0, "mean": 41.1, "n": 21 }, "temperature_c": { "min": 49.0, "median": 55.0, "max": 56.0, "mean": 54.4, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1932.1, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 138.39, "median": 257.94, "max": 258.7, "mean": 248.13, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7524.0, "median": 7524.0, "max": 7524.0, "mean": 7524.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 48.0, "max": 54.0, "mean": 48.0, "n": 18 }, "temperature_c": { "min": 53.0, "median": 55.0, "max": 56.0, "mean": 55.1, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1950.0, "max": 1950.0, "mean": 1946.7, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6075.53, "power_mean_w_all_cards": 519.64, "energy_j_per_1k_tokens_all_cards": 11627.89, "watts_per_tok_s": 6.0755, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 65.0, "temp_min_c": 62.0, "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": 69.0, "fan_max_pct": 69.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "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": 271.51, "power_pct_of_limit": 90.5, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.5% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 65.0, "temp_min_c": 62.0, "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": 69.0, "fan_max_pct": 69.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "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": 271.51, "power_pct_of_limit": 90.5, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.5% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 56.0, "temp_min_c": 53.0, "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": 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": 248.13, "power_pct_of_limit": 82.7, "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 (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8523.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7524.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 272.17, "first_token_was_thinking": false, "wall_s": 6.0043, "eval_count": 256, "eval_duration_ns": 5730282000, "decode_tok_s": 44.675, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25492000, "prefill_tok_s": 40130.237, "load_duration_ns": 243616372, "total_duration_ns": 6002236326, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.486, "decode_tok_s_wall_gross": 42.636, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 107.07, "median": 107.09, "max": 107.11, "mean": 107.09, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 273.32, "max_w": 285.94, "median_w": 284.84, "mean_w_whole_window": 250.57, "max_w_whole_window": 285.94, "limit_w": 300.0, "over_idle_w": 166.23, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 107.05, "median": 284.73, "max": 285.94, "mean": 250.57, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 50.0, "max": 54.0, "mean": 43.6, "n": 21 }, "temperature_c": { "min": 56.0, "median": 65.0, "max": 66.0, "mean": 63.9, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 70.0, "median": 70.0, "max": 71.0, "mean": 70.2, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1965.0, "mean": 1930.0, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 151.18, "median": 284.84, "max": 285.94, "mean": 273.32, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8523.0, "median": 8523.0, "max": 8523.0, "mean": 8523.0, "n": 18 }, "utilization_pct": { "min": 47.0, "median": 51.0, "max": 54.0, "mean": 50.8, "n": 18 }, "temperature_c": { "min": 63.0, "median": 65.0, "max": 66.0, "mean": 65.0, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 70.0, "median": 70.0, "max": 71.0, "mean": 70.3, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1942.5, "max": 1965.0, "mean": 1943.3, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.21, "median": 258.3, "max": 259.36, "mean": 227.38, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7524.0, "median": 7524.0, "max": 7524.0, "mean": 7524.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 51.0, "max": 55.0, "mean": 44.0, "n": 21 }, "temperature_c": { "min": 50.0, "median": 56.0, "max": 57.0, "mean": 55.3, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1934.3, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.0476, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0476, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 137.3, "median": 258.44, "max": 259.36, "mean": 248.83, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7524.0, "median": 7524.0, "max": 7524.0, "mean": 7524.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 53.0, "max": 55.0, "mean": 51.3, "n": 18 }, "temperature_c": { "min": 54.0, "median": 56.0, "max": 57.0, "mean": 55.9, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1950.0, "max": 1950.0, "mean": 1949.2, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 0.0556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0556, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6117.97, "power_mean_w_all_cards": 522.15, "energy_j_per_1k_tokens_all_cards": 11687.76, "watts_per_tok_s": 6.118, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 66.0, "temp_min_c": 63.0, "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": 70.3, "fan_max_pct": 71.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1942.5, "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": 273.32, "power_pct_of_limit": 91.1, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 91.1% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 66.0, "temp_min_c": 63.0, "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": 70.3, "fan_max_pct": 71.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1942.5, "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": 273.32, "power_pct_of_limit": 91.1, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 91.1% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "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": 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": 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": 248.83, "power_pct_of_limit": 82.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.0556, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0556, "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": 8523.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7524.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 14.0, "median": 14.0, "max": 14.0, "mean": 14.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 44.667, "median": 44.675, "max": 44.689, "mean": 44.677, "n": 3 }, "ttft_ms": { "min": 272.17, "median": 277.22, "max": 308.47, "mean": 285.95, "n": 3 }, "power_mean_w": { "min": 267.98, "median": 271.51, "max": 273.32, "mean": 270.94, "n": 3 }, "power_mean_w_all_cards": { "min": 510.06, "median": 519.64, "max": 522.15, "mean": 517.28, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 11419.05, "median": 11627.89, "max": 11687.76, "mean": 11578.23, "n": 3 }, "energy_j_per_1k_tokens": { "min": 5999.45, "median": 6075.53, "max": 6117.97, "mean": 6064.32, "n": 3 } }, "finished_utc": "2026-09-17T01:03:07Z" }, { "num_ctx": 16384, "started_utc": "2026-09-17T01:03:07Z", "loads": true, "size_total": 16117766552, "size_vram": 16117766552, "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": 388.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 354.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8911.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7878.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8523.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7524.0 }, "runner_peak_rss_gib": 0.95, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 111601, "rss_kb": 999360, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 2822 2442 113 42435 44217\nSwap: 24575 0 24575", "kv_footprint_bytes": 1170210816, "kv_footprint_gib": 1.09, "runs": [ { "ttft_ms": 316.05, "first_token_was_thinking": false, "wall_s": 6.0431, "eval_count": 256, "eval_duration_ns": 5725097000, "decode_tok_s": 44.715, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 71829000, "prefill_tok_s": 14242.158, "load_duration_ns": 241086682, "total_duration_ns": 6040939534, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.526, "decode_tok_s_wall_gross": 42.363, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 17.25, "median": 17.31, "max": 17.81, "mean": 17.46, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 267.45, "max_w": 286.33, "median_w": 284.44, "mean_w_whole_window": 233.49, "max_w_whole_window": 286.33, "limit_w": 300.0, "over_idle_w": 249.99, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.33, "median": 284.17, "max": 286.33, "mean": 233.49, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 51.0, "max": 53.0, "mean": 43.0, "n": 21 }, "temperature_c": { "min": 51.0, "median": 65.0, "max": 66.0, "mean": 62.9, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 69.0, "median": 69.0, "max": 70.0, "mean": 69.5, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1965.0, "mean": 1823.6, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 123.15, "median": 284.44, "max": 286.33, "mean": 267.45, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 51.0, "max": 53.0, "mean": 50.1, "n": 18 }, "temperature_c": { "min": 62.0, "median": 65.0, "max": 66.0, "mean": 64.6, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 69.0, "median": 69.0, "max": 70.0, "mean": 69.4, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1950.0, "max": 1965.0, "mean": 1947.5, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.07, "median": 255.87, "max": 257.55, "mean": 209.06, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7878.0, "median": 7878.0, "max": 7878.0, "mean": 7878.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 54.0, "mean": 42.6, "n": 21 }, "temperature_c": { "min": 46.0, "median": 55.0, "max": 56.0, "mean": 53.8, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1950.0, "mean": 1858.6, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 72.1, "median": 256.19, "max": 257.55, "mean": 240.4, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7878.0, "median": 7878.0, "max": 7878.0, "mean": 7878.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 50.0, "max": 54.0, "mean": 49.7, "n": 18 }, "temperature_c": { "min": 53.0, "median": 55.0, "max": 56.0, "mean": 54.8, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1950.0, "max": 1950.0, "mean": 1949.2, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 5981.16, "power_mean_w_all_cards": 507.85, "energy_j_per_1k_tokens_all_cards": 11357.38, "watts_per_tok_s": 5.9812, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 66.0, "temp_min_c": 62.0, "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": 69.4, "fan_max_pct": 70.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1965.0, "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": 267.45, "power_pct_of_limit": 89.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "temp_max_c": 66.0, "temp_min_c": 62.0, "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": 69.4, "fan_max_pct": 70.0, "clock_floor_mhz": 1935.0, "clock_max_mhz": 1965.0, "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": 267.45, "power_pct_of_limit": 89.1, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "temp_max_c": 56.0, "temp_min_c": 53.0, "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": 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": 240.4, "power_pct_of_limit": 80.1, "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": 8911.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7878.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 276.44, "first_token_was_thinking": false, "wall_s": 6.0028, "eval_count": 256, "eval_duration_ns": 5724565000, "decode_tok_s": 44.72, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25178000, "prefill_tok_s": 40630.709, "load_duration_ns": 248395405, "total_duration_ns": 6000789776, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.531, "decode_tok_s_wall_gross": 42.647, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 107.12, "median": 107.14, "max": 107.15, "mean": 107.14, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 272.01, "max_w": 284.3, "median_w": 283.11, "mean_w_whole_window": 249.63, "max_w_whole_window": 284.3, "limit_w": 300.0, "over_idle_w": 164.87, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 107.17, "median": 283.03, "max": 284.3, "mean": 249.63, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 52.0, "mean": 41.3, "n": 21 }, "temperature_c": { "min": 57.0, "median": 66.0, "max": 67.0, "mean": 64.8, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 71.0, "median": 71.0, "max": 72.0, "mean": 71.3, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1965.0, "mean": 1927.9, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 154.24, "median": 283.11, "max": 284.3, "mean": 272.01, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 48.0, "max": 52.0, "mean": 48.2, "n": 18 }, "temperature_c": { "min": 63.0, "median": 66.0, "max": 67.0, "mean": 65.8, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 71.0, "median": 71.0, "max": 72.0, "mean": 71.3, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1942.5, "max": 1965.0, "mean": 1940.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.63, "median": 259.43, "max": 260.59, "mean": 226.96, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7878.0, "median": 7878.0, "max": 7878.0, "mean": 7878.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 54.0, "mean": 42.0, "n": 21 }, "temperature_c": { "min": 50.0, "median": 56.0, "max": 57.0, "mean": 55.5, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1935.7, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 130.38, "median": 259.64, "max": 260.59, "mean": 248.23, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7878.0, "median": 7878.0, "max": 7878.0, "mean": 7878.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 49.5, "max": 54.0, "mean": 49.1, "n": 18 }, "temperature_c": { "min": 54.0, "median": 56.0, "max": 57.0, "mean": 56.2, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6082.57, "power_mean_w_all_cards": 520.24, "energy_j_per_1k_tokens_all_cards": 11633.39, "watts_per_tok_s": 6.0825, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 67.0, "temp_min_c": 63.0, "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": 71.3, "fan_max_pct": 72.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1942.5, "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": 272.01, "power_pct_of_limit": 90.7, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.7% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 67.0, "temp_min_c": 63.0, "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": 71.3, "fan_max_pct": 72.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "clock_median_mhz": 1942.5, "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": 272.01, "power_pct_of_limit": 90.7, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.7% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "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": 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": 248.23, "power_pct_of_limit": 82.7, "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 (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8911.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7878.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 285.57, "first_token_was_thinking": false, "wall_s": 6.0125, "eval_count": 256, "eval_duration_ns": 5725234000, "decode_tok_s": 44.714, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25309000, "prefill_tok_s": 40420.404, "load_duration_ns": 257270765, "total_duration_ns": 6010584194, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.526, "decode_tok_s_wall_gross": 42.578, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 107.42, "median": 107.44, "max": 107.46, "mean": 107.44, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 272.34, "max_w": 283.71, "median_w": 283.43, "mean_w_whole_window": 249.88, "max_w_whole_window": 283.71, "limit_w": 300.0, "over_idle_w": 164.9, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 107.43, "median": 283.38, "max": 283.71, "mean": 249.88, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 54.0, "mean": 42.1, "n": 21 }, "temperature_c": { "min": 57.0, "median": 67.0, "max": 68.0, "mean": 65.7, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 73.0, "median": 73.0, "max": 73.0, "mean": 73.0, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1965.0, "mean": 1925.0, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 154.74, "median": 283.43, "max": 283.71, "mean": 272.34, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8911.0, "median": 8911.0, "max": 8911.0, "mean": 8911.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 49.5, "max": 54.0, "mean": 49.1, "n": 18 }, "temperature_c": { "min": 65.0, "median": 67.0, "max": 68.0, "mean": 66.8, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 73.0, "median": 73.0, "max": 73.0, "mean": 73.0, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1935.0, "max": 1950.0, "mean": 1936.7, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 89.91, "median": 260.44, "max": 261.41, "mean": 228.37, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7878.0, "median": 7878.0, "max": 7878.0, "mean": 7878.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 46.0, "max": 55.0, "mean": 41.2, "n": 21 }, "temperature_c": { "min": 51.0, "median": 57.0, "max": 58.0, "mean": 56.2, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1935.7, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 137.72, "median": 260.56, "max": 261.41, "mean": 249.89, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 7878.0, "median": 7878.0, "max": 7878.0, "mean": 7878.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 46.0, "max": 55.0, "mean": 48.1, "n": 18 }, "temperature_c": { "min": 55.0, "median": 57.0, "max": 58.0, "mean": 57.0, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6090.66, "power_mean_w_all_cards": 522.23, "energy_j_per_1k_tokens_all_cards": 11679.25, "watts_per_tok_s": 6.0907, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 68.0, "temp_min_c": 65.0, "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": 73.0, "fan_max_pct": 73.0, "clock_floor_mhz": 1920.0, "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": 272.34, "power_pct_of_limit": 90.8, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.8% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 68.0, "temp_min_c": 65.0, "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": 73.0, "fan_max_pct": 73.0, "clock_floor_mhz": 1920.0, "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": 272.34, "power_pct_of_limit": 90.8, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 90.8% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "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": 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": 249.89, "power_pct_of_limit": 83.3, "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": 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": 8911.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7878.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 44.714, "median": 44.715, "max": 44.72, "mean": 44.716, "n": 3 }, "ttft_ms": { "min": 276.44, "median": 285.57, "max": 316.05, "mean": 292.69, "n": 3 }, "power_mean_w": { "min": 267.45, "median": 272.01, "max": 272.34, "mean": 270.6, "n": 3 }, "power_mean_w_all_cards": { "min": 507.85, "median": 520.24, "max": 522.23, "mean": 516.77, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 11357.38, "median": 11633.39, "max": 11679.25, "mean": 11556.67, "n": 3 }, "energy_j_per_1k_tokens": { "min": 5981.16, "median": 6082.57, "max": 6090.66, "mean": 6051.46, "n": 3 } }, "finished_utc": "2026-09-17T01:04:21Z" }, { "num_ctx": 32768, "started_utc": "2026-09-17T01:04:21Z", "loads": true, "size_total": 17678047640, "size_vram": 17678047640, "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": 778.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 710.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9689.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8588.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 8911.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 7878.0 }, "runner_peak_rss_gib": 1.02, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 112550, "rss_kb": 1065696, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 2894 2370 177 42500 44146\nSwap: 24575 0 24575", "kv_footprint_bytes": 2730491904, "kv_footprint_gib": 2.54, "runs": [ { "ttft_ms": 278.0, "first_token_was_thinking": false, "wall_s": 6.0071, "eval_count": 256, "eval_duration_ns": 5727274000, "decode_tok_s": 44.698, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 34127000, "prefill_tok_s": 29976.265, "load_duration_ns": 240678666, "total_duration_ns": 6005070659, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.51, "decode_tok_s_wall_gross": 42.616, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 107.39, "median": 107.4, "max": 107.43, "mean": 107.41, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 273.78, "max_w": 285.0, "median_w": 284.11, "mean_w_whole_window": 251.2, "max_w_whole_window": 285.0, "limit_w": 300.0, "over_idle_w": 166.37, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 107.35, "median": 283.83, "max": 285.0, "mean": 251.2, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 46.0, "max": 54.0, "mean": 41.3, "n": 21 }, "temperature_c": { "min": 58.0, "median": 68.0, "max": 69.0, "mean": 66.7, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 73.0, "median": 74.0, "max": 74.0, "mean": 74.0, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1950.0, "mean": 1922.9, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 184.85, "median": 284.11, "max": 285.0, "mean": 273.78, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 47.5, "max": 54.0, "mean": 48.2, "n": 18 }, "temperature_c": { "min": 66.0, "median": 68.0, "max": 69.0, "mean": 67.7, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 73.0, "median": 74.0, "max": 74.0, "mean": 73.9, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1935.0, "max": 1950.0, "mean": 1935.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.14, "median": 260.87, "max": 262.54, "mean": 227.39, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8588.0, "median": 8588.0, "max": 8588.0, "mean": 8588.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 46.0, "max": 54.0, "mean": 41.0, "n": 21 }, "temperature_c": { "min": 51.0, "median": 57.0, "max": 59.0, "mean": 56.7, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1926.4, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 128.6, "median": 261.14, "max": 262.54, "mean": 249.16, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8588.0, "median": 8588.0, "max": 8588.0, "mean": 8588.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 46.0, "max": 54.0, "mean": 47.8, "n": 18 }, "temperature_c": { "min": 56.0, "median": 58.0, "max": 59.0, "mean": 57.4, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1950.0, "max": 1950.0, "mean": 1947.5, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6125.05, "power_mean_w_all_cards": 522.94, "energy_j_per_1k_tokens_all_cards": 11699.3, "watts_per_tok_s": 6.1251, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 69.0, "temp_min_c": 66.0, "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": 73.9, "fan_max_pct": 74.0, "clock_floor_mhz": 1920.0, "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": 273.78, "power_pct_of_limit": 91.3, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 91.3% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 69.0, "temp_min_c": 66.0, "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": 73.9, "fan_max_pct": 74.0, "clock_floor_mhz": 1920.0, "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": 273.78, "power_pct_of_limit": 91.3, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 91.3% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 59.0, "temp_min_c": 56.0, "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": 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": 249.16, "power_pct_of_limit": 83.1, "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": 9689.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8588.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 274.13, "first_token_was_thinking": false, "wall_s": 6.0032, "eval_count": 256, "eval_duration_ns": 5727186000, "decode_tok_s": 44.699, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25262000, "prefill_tok_s": 40495.606, "load_duration_ns": 245743340, "total_duration_ns": 6000994738, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.51, "decode_tok_s_wall_gross": 42.644, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 107.52, "median": 107.52, "max": 107.52, "mean": 107.52, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 276.03, "max_w": 287.87, "median_w": 286.56, "mean_w_whole_window": 253.22, "max_w_whole_window": 287.87, "limit_w": 300.0, "over_idle_w": 168.51, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 107.51, "median": 286.15, "max": 287.87, "mean": 253.22, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 50.0, "max": 54.0, "mean": 42.8, "n": 21 }, "temperature_c": { "min": 59.0, "median": 69.0, "max": 70.0, "mean": 67.5, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 75.0, "median": 75.0, "max": 75.0, "mean": 75.0, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1920.0, "max": 1935.0, "mean": 1915.7, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 157.83, "median": 286.56, "max": 287.87, "mean": 276.03, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 50.0, "max": 54.0, "mean": 49.9, "n": 18 }, "temperature_c": { "min": 67.0, "median": 69.0, "max": 70.0, "mean": 68.5, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 75.0, "median": 75.0, "max": 75.0, "mean": 75.0, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1927.5, "max": 1935.0, "mean": 1927.5, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.47, "median": 261.83, "max": 265.64, "mean": 230.04, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8588.0, "median": 8588.0, "max": 8588.0, "mean": 8588.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 46.0, "max": 55.0, "mean": 41.6, "n": 21 }, "temperature_c": { "min": 52.0, "median": 58.0, "max": 59.0, "mean": 57.3, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1950.0, "mean": 1929.3, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 137.29, "median": 262.05, "max": 265.64, "mean": 251.89, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8588.0, "median": 8588.0, "max": 8588.0, "mean": 8588.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 46.0, "max": 55.0, "mean": 48.6, "n": 18 }, "temperature_c": { "min": 56.0, "median": 58.0, "max": 59.0, "mean": 58.0, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1942.5, "max": 1950.0, "mean": 1942.5, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6175.29, "power_mean_w_all_cards": 527.92, "energy_j_per_1k_tokens_all_cards": 11810.53, "watts_per_tok_s": 6.1753, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 70.0, "temp_min_c": 67.0, "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": 75.0, "fan_max_pct": 75.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1935.0, "clock_median_mhz": 1927.5, "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": 276.03, "power_pct_of_limit": 92.0, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 92.0% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 70.0, "temp_min_c": 67.0, "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": 75.0, "fan_max_pct": 75.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1935.0, "clock_median_mhz": 1927.5, "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": 276.03, "power_pct_of_limit": 92.0, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 92.0% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 59.0, "temp_min_c": 56.0, "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": 1942.5, "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": 251.89, "power_pct_of_limit": 84.0, "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": "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": 9689.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8588.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 273.41, "first_token_was_thinking": false, "wall_s": 6.0035, "eval_count": 256, "eval_duration_ns": 5728299000, "decode_tok_s": 44.69, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 25397000, "prefill_tok_s": 40280.348, "load_duration_ns": 244895163, "total_duration_ns": 6001444818, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.502, "decode_tok_s_wall_gross": 42.642, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 107.62, "median": 107.64, "max": 107.64, "mean": 107.63, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 275.72, "max_w": 288.06, "median_w": 286.35, "mean_w_whole_window": 252.96, "max_w_whole_window": 288.06, "limit_w": 300.0, "over_idle_w": 168.09, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 107.61, "median": 285.99, "max": 288.06, "mean": 252.96, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 46.0, "max": 55.0, "mean": 41.3, "n": 21 }, "temperature_c": { "min": 60.0, "median": 69.0, "max": 70.0, "mean": 68.0, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 76.0, "median": 76.0, "max": 76.0, "mean": 76.0, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1920.0, "max": 1950.0, "mean": 1914.3, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 157.88, "median": 286.35, "max": 288.06, "mean": 275.72, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9689.0, "median": 9689.0, "max": 9689.0, "mean": 9689.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 47.0, "max": 55.0, "mean": 48.2, "n": 18 }, "temperature_c": { "min": 67.0, "median": 69.0, "max": 70.0, "mean": 69.1, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 76.0, "median": 76.0, "max": 76.0, "mean": 76.0, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1905.0, "median": 1920.0, "max": 1935.0, "mean": 1925.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 90.63, "median": 262.88, "max": 266.44, "mean": 230.98, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8588.0, "median": 8588.0, "max": 8588.0, "mean": 8588.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 53.0, "mean": 41.6, "n": 21 }, "temperature_c": { "min": 53.0, "median": 58.0, "max": 60.0, "mean": 57.8, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1935.0, "max": 1950.0, "mean": 1927.9, "n": 21 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 136.45, "median": 263.92, "max": 266.44, "mean": 252.95, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 8588.0, "median": 8588.0, "max": 8588.0, "mean": 8588.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 48.0, "max": 53.0, "mean": 48.6, "n": 18 }, "temperature_c": { "min": 57.0, "median": 58.5, "max": 60.0, "mean": 58.5, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1935.0, "median": 1935.0, "max": 1950.0, "mean": 1940.8, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 6169.56, "power_mean_w_all_cards": 528.67, "energy_j_per_1k_tokens_all_cards": 11829.61, "watts_per_tok_s": 6.1696, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 70.0, "temp_min_c": 67.0, "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": 76.0, "fan_max_pct": 76.0, "clock_floor_mhz": 1905.0, "clock_max_mhz": 1935.0, "clock_median_mhz": 1920.0, "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": 275.72, "power_pct_of_limit": 91.9, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 91.9% of the 300 W cap)" }, "thermal_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 70.0, "temp_min_c": 67.0, "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": 76.0, "fan_max_pct": 76.0, "clock_floor_mhz": 1905.0, "clock_max_mhz": 1935.0, "clock_median_mhz": 1920.0, "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": 275.72, "power_pct_of_limit": 91.9, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "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 (draw held at 91.9% of the 300 W cap)" }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "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": 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": 252.95, "power_pct_of_limit": 84.3, "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": 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": 9689.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8588.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 44.69, "median": 44.698, "max": 44.699, "mean": 44.696, "n": 3 }, "ttft_ms": { "min": 273.41, "median": 274.13, "max": 278.0, "mean": 275.18, "n": 3 }, "power_mean_w": { "min": 273.78, "median": 275.72, "max": 276.03, "mean": 275.18, "n": 3 }, "power_mean_w_all_cards": { "min": 522.94, "median": 527.92, "max": 528.67, "mean": 526.51, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 11699.3, "median": 11810.53, "max": 11829.61, "mean": 11779.81, "n": 3 }, "energy_j_per_1k_tokens": { "min": 6125.05, "median": 6169.56, "max": 6175.29, "mean": 6156.63, "n": 3 } }, "finished_utc": "2026-09-17T01:05:26Z" }, { "num_ctx": 49152, "started_utc": "2026-09-17T01:05:26Z", "loads": true, "size_total": 19189444115, "size_vram": 19189444115, "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": -86.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 718.0 }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9603.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9306.0 }, "memory_used_mib_per_card_before": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9689.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 8588.0 }, "runner_peak_rss_gib": 1.86, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 113521, "rss_kb": 1954708, "threads": 29 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3950 1308 241 42570 43090\nSwap: 24575 0 24575", "kv_footprint_bytes": 4241888379, "kv_footprint_gib": 3.95, "runs": [ { "ttft_ms": 309.01, "first_token_was_thinking": false, "wall_s": 6.0375, "eval_count": 256, "eval_duration_ns": 5726825000, "decode_tok_s": 44.702, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 63343000, "prefill_tok_s": 16150.167, "load_duration_ns": 242771820, "total_duration_ns": 6035618535, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.514, "decode_tok_s_wall_gross": 42.401, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 17.35, "median": 18.17, "max": 18.18, "mean": 17.96, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 19, "mean_w": 259.33, "max_w": 286.91, "median_w": 285.2, "mean_w_whole_window": 236.31, "max_w_whole_window": 286.91, "limit_w": 300.0, "over_idle_w": 241.37, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.35, "median": 284.95, "max": 286.91, "mean": 236.31, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 47.0, "max": 54.0, "mean": 39.9, "n": 21 }, "temperature_c": { "min": 54.0, "median": 68.0, "max": 69.0, "mean": 66.1, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 73.0, "median": 73.0, "max": 74.0, "mean": 73.2, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1950.0, "mean": 1830.7, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.9048, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 54.56, "median": 285.2, "max": 286.91, "mean": 259.33, "n": 19 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 19 }, "utilization_pct": { "min": 25.0, "median": 47.0, "max": 54.0, "mean": 44.1, "n": 19 }, "temperature_c": { "min": 62.0, "median": 68.0, "max": 69.0, "mean": 67.3, "n": 19 }, "memory_temperature_c": null, "fan_pct": { "min": 73.0, "median": 73.0, "max": 74.0, "mean": 73.1, "n": 19 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 19 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1935.0, "max": 1950.0, "mean": 1931.1, "n": 19 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 19 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 19 }, "busy_samples": 19 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.76, "median": 258.55, "max": 259.95, "mean": 211.37, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9306.0, "median": 9306.0, "max": 9306.0, "mean": 9306.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 64.0, "mean": 45.5, "n": 21 }, "temperature_c": { "min": 48.0, "median": 56.0, "max": 58.0, "mean": 55.5, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1950.0, "mean": 1847.9, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 49.42, "median": 258.73, "max": 259.95, "mean": 232.19, "n": 19 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9306.0, "median": 9306.0, "max": 9306.0, "mean": 9306.0, "n": 19 }, "utilization_pct": { "min": 44.0, "median": 48.0, "max": 64.0, "mean": 50.3, "n": 19 }, "temperature_c": { "min": 52.0, "median": 57.0, "max": 58.0, "mean": 56.3, "n": 19 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 19 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 19 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1950.0, "max": 1950.0, "mean": 1942.1, "n": 19 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 19 }, "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": 19 }, "busy_samples": 19 } } }, "energy_j_per_1k_tokens": 5801.32, "power_mean_w_all_cards": 491.52, "energy_j_per_1k_tokens_all_cards": 10995.5, "watts_per_tok_s": 5.8013, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 19, "temp_max_c": 69.0, "temp_min_c": 62.0, "temp_rise_c": 7.0, "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": 73.1, "fan_max_pct": 74.0, "clock_floor_mhz": 1920.0, "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": 259.33, "power_pct_of_limit": 86.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 19, "temp_max_c": 69.0, "temp_min_c": 62.0, "temp_rise_c": 7.0, "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": 73.1, "fan_max_pct": 74.0, "clock_floor_mhz": 1920.0, "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": 259.33, "power_pct_of_limit": 86.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 19, "temp_max_c": 58.0, "temp_min_c": 52.0, "temp_rise_c": 6.0, "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": 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": 232.19, "power_pct_of_limit": 77.4, "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 77.4% 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 (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": 9603.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9306.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 328.55, "first_token_was_thinking": false, "wall_s": 6.0575, "eval_count": 256, "eval_duration_ns": 5727183000, "decode_tok_s": 44.699, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 69809000, "prefill_tok_s": 14654.271, "load_duration_ns": 255864704, "total_duration_ns": 6055499417, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.511, "decode_tok_s_wall_gross": 42.261, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 17.39, "median": 17.45, "max": 17.63, "mean": 17.49, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 266.25, "max_w": 285.58, "median_w": 284.71, "mean_w_whole_window": 232.26, "max_w_whole_window": 285.58, "limit_w": 300.0, "over_idle_w": 248.76, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.36, "median": 284.48, "max": 285.58, "mean": 232.26, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 52.0, "mean": 41.5, "n": 21 }, "temperature_c": { "min": 53.0, "median": 67.0, "max": 69.0, "mean": 65.3, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 72.0, "median": 72.0, "max": 73.0, "mean": 72.4, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1950.0, "mean": 1810.0, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.9524, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 78.62, "median": 284.71, "max": 285.58, "mean": 266.25, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 49.5, "max": 52.0, "mean": 48.4, "n": 18 }, "temperature_c": { "min": 64.0, "median": 67.0, "max": 69.0, "mean": 66.9, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 72.0, "median": 72.0, "max": 73.0, "mean": 72.3, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1935.0, "max": 1950.0, "mean": 1930.8, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.53, "median": 257.92, "max": 259.25, "mean": 211.42, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9306.0, "median": 9306.0, "max": 9306.0, "mean": 9306.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 53.0, "mean": 41.9, "n": 21 }, "temperature_c": { "min": 47.0, "median": 56.0, "max": 57.0, "mean": 54.9, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1950.0, "mean": 1859.3, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.0952, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 76.45, "median": 257.93, "max": 259.25, "mean": 242.92, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9306.0, "median": 9306.0, "max": 9306.0, "mean": 9306.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 49.0, "max": 53.0, "mean": 48.9, "n": 18 }, "temperature_c": { "min": 53.0, "median": 56.0, "max": 57.0, "mean": 55.9, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 0.1111, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 5956.49, "power_mean_w_all_cards": 509.17, "energy_j_per_1k_tokens_all_cards": 11391.05, "watts_per_tok_s": 5.9565, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 69.0, "temp_min_c": 64.0, "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": 72.3, "fan_max_pct": 73.0, "clock_floor_mhz": 1920.0, "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": 266.25, "power_pct_of_limit": 88.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "temp_max_c": 69.0, "temp_min_c": 64.0, "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": 72.3, "fan_max_pct": 73.0, "clock_floor_mhz": 1920.0, "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": 266.25, "power_pct_of_limit": 88.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "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": 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": 242.92, "power_pct_of_limit": 81.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.1111, "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": 9603.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9306.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 318.59, "first_token_was_thinking": false, "wall_s": 6.0465, "eval_count": 256, "eval_duration_ns": 5726138000, "decode_tok_s": 44.707, "prompt_eval_count": 1023, "prompt_eval_duration_ns": 71294000, "prefill_tok_s": 14349.034, "load_duration_ns": 244219393, "total_duration_ns": 6044416434, "streamed_chunks": 256, "response_chars": 1036, "thinking_chars": 0, "done_reason": "length", "num_gpu_option": 99, "decode_tok_s_wall": 44.519, "decode_tok_s_wall_gross": 42.338, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 17.21, "median": 17.29, "max": 17.36, "mean": 17.29, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 6 } }, "measured_over": "samples where the GPU was busy", "busy_samples": 18, "mean_w": 266.52, "max_w": 283.95, "median_w": 283.01, "mean_w_whole_window": 232.59, "max_w_whole_window": 283.95, "limit_w": 300.0, "over_idle_w": 249.23, "samples": 21, "per_gpu": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 17.42, "median": 282.89, "max": 283.95, "mean": 232.59, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 49.0, "max": 55.0, "mean": 43.1, "n": 21 }, "temperature_c": { "min": 53.0, "median": 67.0, "max": 68.0, "mean": 64.9, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 72.0, "median": 72.0, "max": 73.0, "mean": 72.1, "n": 21 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1935.0, "max": 1965.0, "mean": 1802.1, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 21, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 86.82, "median": 283.01, "max": 283.95, "mean": 266.52, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9603.0, "median": 9603.0, "max": 9603.0, "mean": 9603.0, "n": 18 }, "utilization_pct": { "min": 45.0, "median": 51.0, "max": 55.0, "mean": 50.3, "n": 18 }, "temperature_c": { "min": 64.0, "median": 67.0, "max": 68.0, "mean": 66.5, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 72.0, "median": 72.0, "max": 72.0, "mean": 72.0, "n": 18 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 18 }, "clocks_sm_mhz": { "min": 1920.0, "median": 1935.0, "max": 1965.0, "mean": 1935.8, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "throttle_sw_power_cap_fraction": 1.0, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 18 }, "busy_samples": 18 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 11.09, "median": 257.09, "max": 258.77, "mean": 211.25, "n": 21 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9306.0, "median": 9306.0, "max": 9306.0, "mean": 9306.0, "n": 21 }, "utilization_pct": { "min": 0.0, "median": 48.0, "max": 54.0, "mean": 41.8, "n": 21 }, "temperature_c": { "min": 47.0, "median": 56.0, "max": 57.0, "mean": 54.7, "n": 21 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 21 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 }, "clocks_sm_mhz": { "min": 210.0, "median": 1950.0, "max": 1950.0, "mean": 1859.3, "n": 21 }, "clocks_mem_mhz": { "min": 405.0, "median": 9251.0, "max": 9251.0, "mean": 8829.8, "n": 21 }, "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": 21, "busy": { "index": "1", "name": "NVIDIA GeForce RTX 3080", "power_w": { "min": 79.8, "median": 257.49, "max": 258.77, "mean": 242.49, "n": 18 }, "power_limit_w": 300.0, "memory_used_mib": { "min": 9306.0, "median": 9306.0, "max": 9306.0, "mean": 9306.0, "n": 18 }, "utilization_pct": { "min": 44.0, "median": 49.0, "max": 54.0, "mean": 48.8, "n": 18 }, "temperature_c": { "min": 54.0, "median": 56.0, "max": 57.0, "mean": 55.6, "n": 18 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 18 }, "pcie_width_current": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 18 }, "clocks_sm_mhz": { "min": 1950.0, "median": 1950.0, "max": 1950.0, "mean": 1950.0, "n": 18 }, "clocks_mem_mhz": { "min": 9251.0, "median": 9251.0, "max": 9251.0, "mean": 9251.0, "n": 18 }, "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": 18 }, "busy_samples": 18 } } }, "energy_j_per_1k_tokens": 5961.45, "power_mean_w_all_cards": 509.01, "energy_j_per_1k_tokens_all_cards": 11385.4, "watts_per_tok_s": 5.9615, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 18, "temp_max_c": 68.0, "temp_min_c": 64.0, "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": 72.0, "fan_max_pct": 72.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "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": 266.52, "power_pct_of_limit": 88.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "temp_max_c": 68.0, "temp_min_c": 64.0, "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": 72.0, "fan_max_pct": 72.0, "clock_floor_mhz": 1920.0, "clock_max_mhz": 1965.0, "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": 266.52, "power_pct_of_limit": 88.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "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": 18, "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": 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": 242.49, "power_pct_of_limit": 80.8, "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": "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": 9603.0, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": 9306.0 }, "pcie_width_under_load": { "GPU-564e9594-e750-f74a-a9fe-f24c47b8bd4b": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 21 }, "GPU-46e03ea9-af32-5a79-21c9-e59c4413e962": { "min": 4.0, "median": 4.0, "max": 4.0, "mean": 4.0, "n": 21 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 3, "load_pct": { "min": 12.0, "median": 12.0, "max": 12.0, "mean": 12.0, "n": 3 }, "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 44.699, "median": 44.702, "max": 44.707, "mean": 44.703, "n": 3 }, "ttft_ms": { "min": 309.01, "median": 318.59, "max": 328.55, "mean": 318.72, "n": 3 }, "power_mean_w": { "min": 259.33, "median": 266.25, "max": 266.52, "mean": 264.03, "n": 3 }, "power_mean_w_all_cards": { "min": 491.52, "median": 509.01, "max": 509.17, "mean": 503.23, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 10995.5, "median": 11385.4, "max": 11391.05, "mean": 11257.32, "n": 3 }, "energy_j_per_1k_tokens": { "min": 5801.32, "median": 5956.49, "max": 5961.45, "mean": 5906.42, "n": 3 } }, "finished_utc": "2026-09-17T01:07:06Z" }, { "num_ctx": 65536, "started_utc": "2026-09-17T01:07:06Z", "loads": false, "error": "HTTP 500: {\"error\":\"llama-server startup failed after projector CPU offload retry: llama-server process has terminated: exit status 1: cudaMalloc failed: out of memory\\nggml_gallocr_reserve_n_impl: failed to allocate CUDA0 buffer of size 616931328\\ngraph_reserve: failed to allocate compute buffers\\nsched_reserve: compute buffer allocation failed, retrying without pipeline parallelism\\ncudaMalloc f", "verdict": "DOES NOT LOAD at num_ctx 65,536: HTTP 500: {\"error\":\"llama-server startup failed after projector CPU offload retry: llama-server process has terminated: exit status 1: cudaMalloc failed: out of" }, { "num_ctx": 98304, "started_utc": "2026-09-17T01:07:18Z", "loads": false, "error": "HTTP 500: {\"error\":\"llama-server startup failed after projector CPU offload retry: llama-server process has terminated: exit status 1: cudaMalloc failed: out of memory\\nalloc_tensor_range: failed to allocate CUDA0 buffer of size 4492099584\\nllama_init_from_model: failed to initialize the context: failed to allocate buffer for kv cache\"}", "verdict": "DOES NOT LOAD at num_ctx 98,304: HTTP 500: {\"error\":\"llama-server startup failed after projector CPU offload retry: llama-server process has terminated: exit status 1: cudaMalloc failed: out of" }, { "num_ctx": 131072, "started_utc": "2026-09-17T01:07:28Z", "loads": false, "error": "HTTP 500: {\"error\":\"llama-server startup failed after projector CPU offload retry: llama-server process has terminated: exit status 1: cudaMalloc failed: out of memory\\nalloc_tensor_range: failed to allocate CUDA0 buffer of size 5989466112\\nllama_init_from_model: failed to initialize the context: failed to allocate buffer for kv cache\"}", "verdict": "DOES NOT LOAD at num_ctx 131,072: HTTP 500: {\"error\":\"llama-server startup failed after projector CPU offload retry: llama-server process has terminated: exit status 1: cudaMalloc failed: out of" } ], "errors": [], "context_length_max": 131072, "quantization": "Q4_K_M", "ups_after_arm": { "ups": "pr1500@localhost", "read_utc": "2026-09-17T01:07:37Z", "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: 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": 122.0, "output.voltage": 122.0, "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": 49152, "levels_requested": [ 4096, 8192, 16384, 32768, 49152, 65536, 98304, 131072 ], "levels_not_reached": [], "headline": "the biggest context this arm holds for mistral-small3.2:24b is 49,152 tokens", "finished_utc": "2026-09-17T01:07:38Z" }