{ "model_id": "m5", "model_tag": "gemma4:26b", "arm": "one", "box_class": "desktop-3090", "mode": "context-ladder", "arm_cards": [ "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08" ], "target_uuid": "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08", "started_utc": "2026-09-18T00:22:57Z", "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 card, 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 card's window.", "pcie_note": "This card negotiated PCIe x16 of a x16-capable generation-3 link -- the x16 slot of the same board on which the pair bench measured a x16 + x4 asymmetry. With one card there is no per-token traffic BETWEEN cards, so the link is not in the decode path the way it was for the split pair; it still carries the model in and the tokens out. Every run records `pcie.link.width.current` 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": "ONE GeForce RTX 3090 24 GB in a consumer desktop, alone on the board's x16 slot -- the same box, the same case and the same airflow in which two GeForce RTX 3080 10 GB cards were measured the night before. A temperature here is a reading of THAT arrangement. It differs from the pair's arrangement in the way that matters most to a temperature: with one card there is no second board warming the air or blocking a face, so this card has open air on both sides where the pair did not. The card is traced on every run.", "card_labels": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": "index 0 - EVGA - GPU-7aa0be10" }, "card_identity": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": 0, "bus_id": "00000000:01:00.0", "pci_sub_device_id": "0x39753842", "vendor": "EVGA", "board_per_operator": "EVGA GeForce RTX 3090 XC3 Ultra 24 GB", "vbios": "94.02.42.80.C1", "serial": null, "pcie_link_width_max": 16, "pcie_link_width_negotiated": 16, "memory_total_mib": 24576, "power_default_limit_w": 350.0, "power_min_limit_w": 100.0, "power_max_limit_w": 366.0, "power_limit_at_bench_start_w": 250.0, "note": "the single card seated in this box's x16 slot for the control arm. It came up at 280 W from a boot unit written for the box's former pair (/etc/desktop/gpu-power-caps.conf, 2026-08-27) whose other row names a card that is no longer present; the operator set 250 W by hand before this bench and the read-back is recorded per arm." } }, "memory_temp_support": { "checked_utc": "2026-09-18T00:22:57Z", "paths": { "nvidia-smi --query-gpu=temperature.memory": { "raw": "0, GPU-7aa0be10-974f-6430-52fe-09018a0e2e08, 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" ], "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 } } } }, "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": null, "num_gpu_policy": "layers left to ollama's own planner (`auto`)", "num_gpu_note": "ollama decides for itself how many of a model's layers to put on the card. On the pair of 3080s that decision was measurably conservative -- 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 -- so the pair bench reported `auto` and a forced layer count side by side. On ONE 24 GB card there is no placement decision to make: the expected reading is 100% on the card with no spill. This harness records `options.num_gpu` on every run as `num_gpu_option` and reads the planner's actual placement back from /api/ps, so a spill on a card with room is visible as a finding rather than absorbed into a tok/s figure. A value at or above the model's layer count means every layer.", "instance_env": { "written_utc": "2026-09-18T00:22:49Z", "unit": "bench-3090-one", "shape": "one", "port": 11470, "base_url": "http://127.0.0.1:11470", "cuda_visible_devices": "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08", "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, 350.00 W;", "persistence_mode_at_start": "0, Enabled;", "pcie_link_width_at_start": "0, 16;" }, "ups_before_arm": { "ups": "pr1500@localhost", "read_utc": "2026-09-18T00:22:57Z", "available": false, "error": "upsc rc=1: Error: Connection failure: Connection refused" }, "levels": [ { "num_ctx": 4096, "started_utc": "2026-09-18T00:22:58Z", "loads": true, "size_total": 17201018960, "size_vram": 17201018960, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 17658.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 559.0 }, "runner_peak_rss_gib": 1.28, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 6741, "rss_kb": 1346792, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3532 22319 44 21755 43508\nSwap: 24575 0 24575", "kv_footprint_bytes": 0, "kv_footprint_gib": 0.0, "runs": [ { "ttft_ms": 592.12, "first_token_was_thinking": false, "wall_s": 2.4473, "eval_count": 256, "eval_duration_ns": 1852160000, "decode_tok_s": 138.217, "prompt_eval_count": 536, "prompt_eval_duration_ns": 132128999, "prefill_tok_s": 4056.642, "load_duration_ns": 455620159, "total_duration_ns": 2444009228, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 137.449, "decode_tok_s_wall_gross": 104.603, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 20.04, "median": 20.18, "max": 20.33, "mean": 20.18, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20.18 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "power_w_all_cards": 20.18, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 214.86, "max_w": 319.52, "median_w": 231.34, "mean_w_whole_window": 150.04, "max_w_whole_window": 319.52, "limit_w": 350.0, "over_idle_w": 194.68, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 20.15, "median": 140.59, "max": 319.52, "mean": 150.04, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 36.0, "max": 91.0, "mean": 42.3, "n": 9 }, "temperature_c": { "min": 31.0, "median": 39.0, "max": 42.0, "mean": 37.0, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1920.0, "max": 1920.0, "mean": 1511.7, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 50.95, "median": 231.34, "max": 319.52, "mean": 214.86, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 6 }, "utilization_pct": { "min": 36.0, "median": 63.5, "max": 91.0, "mean": 63.5, "n": 6 }, "temperature_c": { "min": 36.0, "median": 40.5, "max": 42.0, "mean": 40.0, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1920.0, "max": 1920.0, "mean": 1910.0, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.8333, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1554.51, "power_mean_w_all_cards": 214.86, "energy_j_per_1k_tokens_all_cards": 1554.51, "watts_per_tok_s": 1.5545, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 42.0, "temp_min_c": 36.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1920.0, "clock_median_mhz": 1920.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 214.86, "power_pct_of_limit": 61.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8333, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 42.0, "temp_min_c": 36.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1920.0, "clock_median_mhz": 1920.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 214.86, "power_pct_of_limit": 61.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8333, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 564.53, "first_token_was_thinking": false, "wall_s": 2.4192, "eval_count": 256, "eval_duration_ns": 1851713000, "decode_tok_s": 138.25, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102066000, "prefill_tok_s": 5251.504, "load_duration_ns": 458240914, "total_duration_ns": 2415925083, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 137.491, "decode_tok_s_wall_gross": 105.821, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 19.99, "median": 20.49, "max": 21.3, "mean": 20.59, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20.59 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "power_w_all_cards": 20.59, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 219.02, "max_w": 318.92, "median_w": 238.59, "mean_w_whole_window": 153.41, "max_w_whole_window": 318.92, "limit_w": 350.0, "over_idle_w": 198.43, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.29, "median": 141.02, "max": 318.92, "mean": 153.41, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 77.0, "max": 92.0, "mean": 58.0, "n": 9 }, "temperature_c": { "min": 33.0, "median": 42.0, "max": 45.0, "mean": 39.8, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1905.0, "max": 1905.0, "mean": 1503.3, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 59.87, "median": 238.59, "max": 318.92, "mean": 219.02, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 6 }, "utilization_pct": { "min": 77.0, "median": 92.0, "max": 92.0, "mean": 87.0, "n": 6 }, "temperature_c": { "min": 38.0, "median": 43.5, "max": 45.0, "mean": 42.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1905.0, "max": 1905.0, "mean": 1897.5, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1584.23, "power_mean_w_all_cards": 219.02, "energy_j_per_1k_tokens_all_cards": 1584.23, "watts_per_tok_s": 1.5842, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 45.0, "temp_min_c": 38.0, "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": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1905.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 219.02, "power_pct_of_limit": 62.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 45.0, "temp_min_c": 38.0, "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": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1905.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 219.02, "power_pct_of_limit": 62.6, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 547.08, "first_token_was_thinking": false, "wall_s": 2.4078, "eval_count": 256, "eval_duration_ns": 1857760000, "decode_tok_s": 137.8, "prompt_eval_count": 536, "prompt_eval_duration_ns": 80405000, "prefill_tok_s": 6666.252, "load_duration_ns": 462437534, "total_duration_ns": 2404533828, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 137.046, "decode_tok_s_wall_gross": 106.322, "num_ctx_option": 4096, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 114.99, "median": 115.18, "max": 115.4, "mean": 115.19, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 115.19 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "power_w_all_cards": 115.19, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 245.83, "max_w": 315.64, "median_w": 258.06, "mean_w_whole_window": 202.41, "max_w_whole_window": 315.64, "limit_w": 350.0, "over_idle_w": 130.64, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 115.55, "median": 190.11, "max": 315.64, "mean": 202.41, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 61.3, "n": 9 }, "temperature_c": { "min": 39.0, "median": 45.0, "max": 48.0, "mean": 43.6, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1890.0, "max": 1890.0, "mean": 1833.3, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 138.02, "median": 258.06, "max": 315.64, "mean": 245.83, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18217.0, "median": 18217.0, "max": 18217.0, "mean": 18217.0, "n": 6 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 6 }, "temperature_c": { "min": 42.0, "median": 46.5, "max": 48.0, "mean": 45.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1890.0, "max": 1890.0, "mean": 1887.5, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1783.96, "power_mean_w_all_cards": 245.83, "energy_j_per_1k_tokens_all_cards": 1783.96, "watts_per_tok_s": 1.784, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 48.0, "temp_min_c": 42.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 245.83, "power_pct_of_limit": 70.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 48.0, "temp_min_c": 42.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 245.83, "power_pct_of_limit": 70.2, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 137.8, "median": 138.217, "max": 138.25, "mean": 138.089, "n": 3 }, "ttft_ms": { "min": 547.08, "median": 564.53, "max": 592.12, "mean": 567.91, "n": 3 }, "power_mean_w": { "min": 214.86, "median": 219.02, "max": 245.83, "mean": 226.57, "n": 3 }, "power_mean_w_all_cards": { "min": 214.86, "median": 219.02, "max": 245.83, "mean": 226.57, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1554.51, "median": 1584.23, "max": 1783.96, "mean": 1640.9, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1554.51, "median": 1584.23, "max": 1783.96, "mean": 1640.9, "n": 3 } }, "finished_utc": "2026-09-18T00:24:17Z" }, { "num_ctx": 8192, "started_utc": "2026-09-18T00:24:17Z", "loads": true, "size_total": 17372607938, "size_vram": 17372607938, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 208.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18217.0 }, "runner_peak_rss_gib": 1.29, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 7726, "rss_kb": 1356956, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3515 22322 70 21794 43525\nSwap: 24575 0 24575", "kv_footprint_bytes": 171588978, "kv_footprint_gib": 0.16, "runs": [ { "ttft_ms": 549.77, "first_token_was_thinking": false, "wall_s": 2.4134, "eval_count": 256, "eval_duration_ns": 1860615000, "decode_tok_s": 137.589, "prompt_eval_count": 536, "prompt_eval_duration_ns": 90006000, "prefill_tok_s": 5955.159, "load_duration_ns": 455530524, "total_duration_ns": 2410116021, "streamed_chunks": 256, "response_chars": 1034, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.832, "decode_tok_s_wall_gross": 106.076, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 117.15, "median": 117.33, "max": 117.35, "mean": 117.28, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 117.28 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "power_w_all_cards": 117.28, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 251.53, "max_w": 319.2, "median_w": 267.8, "mean_w_whole_window": 206.97, "max_w_whole_window": 319.2, "limit_w": 350.0, "over_idle_w": 134.25, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 117.33, "median": 201.27, "max": 319.2, "mean": 206.97, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 90.0, "max": 93.0, "mean": 61.1, "n": 9 }, "temperature_c": { "min": 42.0, "median": 49.0, "max": 51.0, "mean": 46.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1875.0, "max": 1890.0, "mean": 1828.3, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 135.99, "median": 267.8, "max": 319.2, "mean": 251.53, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 6 }, "utilization_pct": { "min": 90.0, "median": 92.0, "max": 93.0, "mean": 91.7, "n": 6 }, "temperature_c": { "min": 46.0, "median": 50.0, "max": 51.0, "mean": 49.3, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1875.0, "max": 1890.0, "mean": 1880.0, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "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": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1828.13, "power_mean_w_all_cards": 251.53, "energy_j_per_1k_tokens_all_cards": 1828.13, "watts_per_tok_s": 1.8281, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 51.0, "temp_min_c": 46.0, "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": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 251.53, "power_pct_of_limit": 71.9, "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 (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 51.0, "temp_min_c": 46.0, "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": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 251.53, "power_pct_of_limit": 71.9, "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 (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 565.98, "first_token_was_thinking": false, "wall_s": 2.4274, "eval_count": 256, "eval_duration_ns": 1858508000, "decode_tok_s": 137.745, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102314000, "prefill_tok_s": 5238.775, "load_duration_ns": 459277144, "total_duration_ns": 2424161188, "streamed_chunks": 256, "response_chars": 1034, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.991, "decode_tok_s_wall_gross": 105.462, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 21.18, "median": 21.42, "max": 22.51, "mean": 21.7, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 21.7 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "power_w_all_cards": 21.7, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 233.65, "max_w": 337.3, "median_w": 253.96, "mean_w_whole_window": 163.07, "max_w_whole_window": 337.3, "limit_w": 350.0, "over_idle_w": 211.95, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.09, "median": 161.19, "max": 337.3, "mean": 163.07, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 61.3, "n": 9 }, "temperature_c": { "min": 40.0, "median": 50.0, "max": 52.0, "mean": 47.4, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1845.0, "max": 1905.0, "mean": 1468.3, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 59.93, "median": 253.96, "max": 337.3, "mean": 233.65, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 6 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 6 }, "temperature_c": { "min": 47.0, "median": 52.0, "max": 52.0, "mean": 50.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1830.0, "median": 1875.0, "max": 1905.0, "mean": 1872.5, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1696.25, "power_mean_w_all_cards": 233.65, "energy_j_per_1k_tokens_all_cards": 1696.25, "watts_per_tok_s": 1.6963, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 52.0, "temp_min_c": 47.0, "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": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 233.65, "power_pct_of_limit": 66.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 52.0, "temp_min_c": 47.0, "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": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 233.65, "power_pct_of_limit": 66.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 565.86, "first_token_was_thinking": false, "wall_s": 2.4346, "eval_count": 256, "eval_duration_ns": 1865825000, "decode_tok_s": 137.205, "prompt_eval_count": 536, "prompt_eval_duration_ns": 103430000, "prefill_tok_s": 5182.249, "load_duration_ns": 458158936, "total_duration_ns": 2431369890, "streamed_chunks": 256, "response_chars": 1034, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.452, "decode_tok_s_wall_gross": 105.149, "num_ctx_option": 8192, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 21.28, "median": 21.34, "max": 21.38, "mean": 21.33, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 21.33 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "power_w_all_cards": 21.33, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 256.54, "max_w": 321.95, "median_w": 294.47, "mean_w_whole_window": 155.94, "max_w_whole_window": 321.95, "limit_w": 350.0, "over_idle_w": 235.21, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.4, "median": 156.02, "max": 321.95, "mean": 155.94, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 51.1, "n": 9 }, "temperature_c": { "min": 42.0, "median": 51.0, "max": 53.0, "mean": 48.6, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1875.0, "max": 1875.0, "mean": 1466.7, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 156.02, "median": 294.47, "max": 321.95, "mean": 256.54, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18425.0, "median": 18425.0, "max": 18425.0, "mean": 18425.0, "n": 5 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 5 }, "temperature_c": { "min": 51.0, "median": 53.0, "max": 53.0, "mean": 52.4, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1875.0, "max": 1875.0, "mean": 1875.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1869.76, "power_mean_w_all_cards": 256.54, "energy_j_per_1k_tokens_all_cards": 1869.76, "watts_per_tok_s": 1.8698, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 53.0, "temp_min_c": 51.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 256.54, "power_pct_of_limit": 73.3, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 53.0, "temp_min_c": 51.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 256.54, "power_pct_of_limit": 73.3, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 137.205, "median": 137.589, "max": 137.745, "mean": 137.513, "n": 3 }, "ttft_ms": { "min": 549.77, "median": 565.86, "max": 565.98, "mean": 560.54, "n": 3 }, "power_mean_w": { "min": 233.65, "median": 251.53, "max": 256.54, "mean": 247.24, "n": 3 }, "power_mean_w_all_cards": { "min": 233.65, "median": 251.53, "max": 256.54, "mean": 247.24, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1696.25, "median": 1828.13, "max": 1869.76, "mean": 1798.05, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1696.25, "median": 1828.13, "max": 1869.76, "mean": 1798.05, "n": 3 } }, "finished_utc": "2026-09-18T00:25:28Z" }, { "num_ctx": 16384, "started_utc": "2026-09-18T00:25:28Z", "loads": true, "size_total": 17239291984, "size_vram": 17239291984, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": -46.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18425.0 }, "runner_peak_rss_gib": 1.3, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 8596, "rss_kb": 1360120, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3438 22398 56 21781 43601\nSwap: 24575 0 24575", "kv_footprint_bytes": 38273024, "kv_footprint_gib": 0.04, "runs": [ { "ttft_ms": 555.0, "first_token_was_thinking": false, "wall_s": 2.4232, "eval_count": 256, "eval_duration_ns": 1865180000, "decode_tok_s": 137.252, "prompt_eval_count": 536, "prompt_eval_duration_ns": 93182000, "prefill_tok_s": 5752.184, "load_duration_ns": 457478511, "total_duration_ns": 2419880173, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.493, "decode_tok_s_wall_gross": 105.644, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 122.1, "median": 122.11, "max": 122.14, "mean": 122.12, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 122.12 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "power_w_all_cards": 122.12, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 276.47, "max_w": 321.81, "median_w": 299.61, "mean_w_whole_window": 210.08, "max_w_whole_window": 321.81, "limit_w": 350.0, "over_idle_w": 154.35, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 122.16, "median": 207.86, "max": 321.81, "mean": 210.08, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 50.9, "n": 9 }, "temperature_c": { "min": 48.0, "median": 54.0, "max": 56.0, "mean": 52.6, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1875.0, "max": 1905.0, "mean": 1833.3, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 207.86, "median": 299.61, "max": 321.81, "mean": 276.47, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.6, "n": 5 }, "temperature_c": { "min": 54.0, "median": 55.0, "max": 56.0, "mean": 55.2, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1890.0, "max": 1905.0, "mean": 1890.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 2014.32, "power_mean_w_all_cards": 276.47, "energy_j_per_1k_tokens_all_cards": 2014.32, "watts_per_tok_s": 2.0143, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 56.0, "temp_min_c": 54.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 276.47, "power_pct_of_limit": 79.0, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 56.0, "temp_min_c": 54.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 276.47, "power_pct_of_limit": 79.0, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 572.46, "first_token_was_thinking": false, "wall_s": 2.4495, "eval_count": 256, "eval_duration_ns": 1874191000, "decode_tok_s": 136.592, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102562000, "prefill_tok_s": 5226.107, "load_duration_ns": 465575639, "total_duration_ns": 2446315780, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.851, "decode_tok_s_wall_gross": 104.51, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 21.3, "median": 21.31, "max": 21.48, "mean": 21.36, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 21.36 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "power_w_all_cards": 21.36, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 249.61, "max_w": 317.01, "median_w": 293.06, "mean_w_whole_window": 152.49, "max_w_whole_window": 317.01, "limit_w": 350.0, "over_idle_w": 228.25, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.24, "median": 150.6, "max": 317.01, "mean": 152.49, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 50.9, "n": 9 }, "temperature_c": { "min": 44.0, "median": 52.0, "max": 55.0, "mean": 50.4, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1875.0, "max": 1890.0, "mean": 1473.3, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 150.6, "median": 293.06, "max": 317.01, "mean": 249.61, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.6, "n": 5 }, "temperature_c": { "min": 52.0, "median": 54.0, "max": 55.0, "mean": 54.0, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1890.0, "max": 1890.0, "mean": 1887.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1827.41, "power_mean_w_all_cards": 249.61, "energy_j_per_1k_tokens_all_cards": 1827.41, "watts_per_tok_s": 1.8274, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 55.0, "temp_min_c": 52.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 249.61, "power_pct_of_limit": 71.3, "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 (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 55.0, "temp_min_c": 52.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 249.61, "power_pct_of_limit": 71.3, "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 (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 565.84, "first_token_was_thinking": false, "wall_s": 2.4396, "eval_count": 256, "eval_duration_ns": 1870641000, "decode_tok_s": 136.851, "prompt_eval_count": 536, "prompt_eval_duration_ns": 103765000, "prefill_tok_s": 5165.518, "load_duration_ns": 457694201, "total_duration_ns": 2436265329, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.089, "decode_tok_s_wall_gross": 104.934, "num_ctx_option": 16384, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 21.51, "median": 21.82, "max": 21.95, "mean": 21.76, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 21.76 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "power_w_all_cards": 21.76, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 256.01, "max_w": 318.58, "median_w": 305.66, "mean_w_whole_window": 155.97, "max_w_whole_window": 318.58, "limit_w": 350.0, "over_idle_w": 234.25, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.58, "median": 152.02, "max": 318.58, "mean": 155.97, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 91.0, "mean": 50.6, "n": 9 }, "temperature_c": { "min": 45.0, "median": 54.0, "max": 56.0, "mean": 51.3, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1890.0, "max": 1890.0, "mean": 1475.0, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.7778, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 152.02, "median": 305.66, "max": 318.58, "mean": 256.01, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18379.0, "median": 18379.0, "max": 18379.0, "mean": 18379.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 91.0, "max": 91.0, "mean": 91.0, "n": 5 }, "temperature_c": { "min": 54.0, "median": 55.0, "max": 56.0, "mean": 55.2, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1890.0, "median": 1890.0, "max": 1890.0, "mean": 1890.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1870.71, "power_mean_w_all_cards": 256.01, "energy_j_per_1k_tokens_all_cards": 1870.71, "watts_per_tok_s": 1.8707, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 56.0, "temp_min_c": 54.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1890.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 256.01, "power_pct_of_limit": 73.1, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 56.0, "temp_min_c": 54.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1890.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 256.01, "power_pct_of_limit": 73.1, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 136.592, "median": 136.851, "max": 137.252, "mean": 136.898, "n": 3 }, "ttft_ms": { "min": 555.0, "median": 565.84, "max": 572.46, "mean": 564.43, "n": 3 }, "power_mean_w": { "min": 249.61, "median": 256.01, "max": 276.47, "mean": 260.7, "n": 3 }, "power_mean_w_all_cards": { "min": 249.61, "median": 256.01, "max": 276.47, "mean": 260.7, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1827.41, "median": 1870.71, "max": 2014.32, "mean": 1904.15, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1827.41, "median": 1870.71, "max": 2014.32, "mean": 1904.15, "n": 3 } }, "finished_utc": "2026-09-18T00:26:50Z" }, { "num_ctx": 32768, "started_utc": "2026-09-18T00:26:50Z", "loads": true, "size_total": 17323178064, "size_vram": 17323178064, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 250.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18379.0 }, "runner_peak_rss_gib": 1.31, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 9642, "rss_kb": 1378492, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3512 22299 72 21822 43527\nSwap: 24575 0 24575", "kv_footprint_bytes": 122159104, "kv_footprint_gib": 0.11, "runs": [ { "ttft_ms": 576.19, "first_token_was_thinking": false, "wall_s": 2.4435, "eval_count": 256, "eval_duration_ns": 1864193000, "decode_tok_s": 137.325, "prompt_eval_count": 536, "prompt_eval_duration_ns": 113711000, "prefill_tok_s": 4713.704, "load_duration_ns": 458100296, "total_duration_ns": 2440115984, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.561, "decode_tok_s_wall_gross": 104.768, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 21.67, "median": 21.92, "max": 22.09, "mean": 21.89, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 21.89 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "power_w_all_cards": 21.89, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 190.71, "max_w": 318.91, "median_w": 181.48, "mean_w_whole_window": 153.17, "max_w_whole_window": 318.91, "limit_w": 350.0, "over_idle_w": 168.82, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.8, "median": 147.32, "max": 318.91, "mean": 153.17, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 59.4, "n": 9 }, "temperature_c": { "min": 44.0, "median": 53.0, "max": 55.0, "mean": 50.6, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1890.0, "max": 1890.0, "mean": 1491.7, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.7778, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 22.81, "median": 181.48, "max": 318.91, "mean": 190.71, "n": 7 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 7 }, "utilization_pct": { "min": 39.0, "median": 91.0, "max": 92.0, "mean": 76.4, "n": 7 }, "temperature_c": { "min": 45.0, "median": 54.0, "max": 55.0, "mean": 52.4, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1890.0, "max": 1890.0, "mean": 1857.9, "n": 7 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 7 }, "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": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1388.75, "power_mean_w_all_cards": 190.71, "energy_j_per_1k_tokens_all_cards": 1388.75, "watts_per_tok_s": 1.3887, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 55.0, "temp_min_c": 45.0, "temp_rise_c": 10.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1725.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 350.0, "power_mean_w": 190.71, "power_pct_of_limit": 54.5, "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": "NEITHER cap: the SM clock varied with draw at 54.5% of the cap and the card at 55 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 55.0, "temp_min_c": 45.0, "temp_rise_c": 10.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1725.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 350.0, "power_mean_w": 190.71, "power_pct_of_limit": 54.5, "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": "NEITHER cap: the SM clock varied with draw at 54.5% of the cap and the card at 55 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 564.33, "first_token_was_thinking": false, "wall_s": 2.425, "eval_count": 256, "eval_duration_ns": 1857768000, "decode_tok_s": 137.8, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102959000, "prefill_tok_s": 5205.956, "load_duration_ns": 456744185, "total_duration_ns": 2421678836, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 137.049, "decode_tok_s_wall_gross": 105.568, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 21.44, "median": 21.64, "max": 21.66, "mean": 21.58, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 21.58 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "power_w_all_cards": 21.58, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 258.5, "max_w": 326.63, "median_w": 308.36, "mean_w_whole_window": 158.44, "max_w_whole_window": 326.63, "limit_w": 350.0, "over_idle_w": 236.92, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 21.84, "median": 148.88, "max": 326.63, "mean": 158.44, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 51.0, "n": 9 }, "temperature_c": { "min": 45.0, "median": 54.0, "max": 57.0, "mean": 51.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1890.0, "max": 1905.0, "mean": 1478.3, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.7778, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 148.88, "median": 308.36, "max": 326.63, "mean": 258.5, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.8, "n": 5 }, "temperature_c": { "min": 54.0, "median": 56.0, "max": 57.0, "mean": 55.6, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1890.0, "median": 1905.0, "max": 1905.0, "mean": 1899.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1875.91, "power_mean_w_all_cards": 258.5, "energy_j_per_1k_tokens_all_cards": 1875.91, "watts_per_tok_s": 1.8759, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 57.0, "temp_min_c": 54.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1890.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1905.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 258.5, "power_pct_of_limit": 73.9, "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 (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 57.0, "temp_min_c": 54.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1890.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1905.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 258.5, "power_pct_of_limit": 73.9, "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 (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 541.46, "first_token_was_thinking": false, "wall_s": 2.4092, "eval_count": 256, "eval_duration_ns": 1864473000, "decode_tok_s": 137.304, "prompt_eval_count": 536, "prompt_eval_duration_ns": 80209000, "prefill_tok_s": 6682.542, "load_duration_ns": 456669739, "total_duration_ns": 2405615411, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.532, "decode_tok_s_wall_gross": 106.261, "num_ctx_option": 32768, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 122.61, "median": 122.61, "max": 122.66, "mean": 122.63, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 122.63 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "power_w_all_cards": 122.63, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 249.8, "max_w": 317.16, "median_w": 261.82, "mean_w_whole_window": 207.5, "max_w_whole_window": 317.16, "limit_w": 350.0, "over_idle_w": 127.17, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 122.69, "median": 196.4, "max": 317.16, "mean": 207.5, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 74.0, "max": 92.0, "mean": 57.3, "n": 9 }, "temperature_c": { "min": 51.0, "median": 56.0, "max": 58.0, "mean": 54.8, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1875.0, "max": 1890.0, "mean": 1825.0, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 145.38, "median": 261.82, "max": 317.16, "mean": 249.8, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18629.0, "median": 18629.0, "max": 18629.0, "mean": 18629.0, "n": 6 }, "utilization_pct": { "min": 74.0, "median": 92.0, "max": 92.0, "mean": 86.0, "n": 6 }, "temperature_c": { "min": 54.0, "median": 57.0, "max": 58.0, "mean": 56.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1875.0, "max": 1890.0, "mean": 1875.0, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1819.32, "power_mean_w_all_cards": 249.8, "energy_j_per_1k_tokens_all_cards": 1819.32, "watts_per_tok_s": 1.8193, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 58.0, "temp_min_c": 54.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 249.8, "power_pct_of_limit": 71.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 58.0, "temp_min_c": 54.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1890.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 249.8, "power_pct_of_limit": 71.4, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 137.304, "median": 137.325, "max": 137.8, "mean": 137.476, "n": 3 }, "ttft_ms": { "min": 541.46, "median": 564.33, "max": 576.19, "mean": 560.66, "n": 3 }, "power_mean_w": { "min": 190.71, "median": 249.8, "max": 258.5, "mean": 233.0, "n": 3 }, "power_mean_w_all_cards": { "min": 190.71, "median": 249.8, "max": 258.5, "mean": 233.0, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1388.75, "median": 1819.32, "max": 1875.91, "mean": 1694.66, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1388.75, "median": 1819.32, "max": 1875.91, "mean": 1694.66, "n": 3 } }, "finished_utc": "2026-09-18T00:28:21Z" }, { "num_ctx": 49152, "started_utc": "2026-09-18T00:28:21Z", "loads": true, "size_total": 17407064144, "size_vram": 17407064144, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 250.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18629.0 }, "runner_peak_rss_gib": 1.33, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 10702, "rss_kb": 1395616, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3487 22324 88 21838 43553\nSwap: 24575 0 24575", "kv_footprint_bytes": 206045184, "kv_footprint_gib": 0.19, "runs": [ { "ttft_ms": 553.99, "first_token_was_thinking": false, "wall_s": 2.4225, "eval_count": 256, "eval_duration_ns": 1865423000, "decode_tok_s": 137.234, "prompt_eval_count": 536, "prompt_eval_duration_ns": 92604000, "prefill_tok_s": 5788.087, "load_duration_ns": 456881002, "total_duration_ns": 2419095723, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.475, "decode_tok_s_wall_gross": 105.677, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 124.82, "median": 124.86, "max": 124.87, "mean": 124.85, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 124.85 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "power_w_all_cards": 124.85, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 251.54, "max_w": 319.16, "median_w": 263.54, "mean_w_whole_window": 209.35, "max_w_whole_window": 319.16, "limit_w": 350.0, "over_idle_w": 126.69, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 124.86, "median": 197.48, "max": 319.16, "mean": 209.35, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 72.0, "max": 92.0, "mean": 56.9, "n": 9 }, "temperature_c": { "min": 53.0, "median": 58.0, "max": 60.0, "mean": 56.8, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1860.0, "max": 1875.0, "mean": 1816.7, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 147.33, "median": 263.54, "max": 319.16, "mean": 251.54, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 6 }, "utilization_pct": { "min": 72.0, "median": 92.0, "max": 92.0, "mean": 85.3, "n": 6 }, "temperature_c": { "min": 56.0, "median": 59.0, "max": 60.0, "mean": 58.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1830.0, "median": 1867.5, "max": 1875.0, "mean": 1862.5, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1832.92, "power_mean_w_all_cards": 251.54, "energy_j_per_1k_tokens_all_cards": 1832.92, "watts_per_tok_s": 1.8329, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 60.0, "temp_min_c": 56.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1867.5, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 251.54, "power_pct_of_limit": 71.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 60.0, "temp_min_c": 56.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1867.5, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 251.54, "power_pct_of_limit": 71.9, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 543.4, "first_token_was_thinking": false, "wall_s": 2.421, "eval_count": 256, "eval_duration_ns": 1874588000, "decode_tok_s": 136.563, "prompt_eval_count": 536, "prompt_eval_duration_ns": 80178000, "prefill_tok_s": 6685.126, "load_duration_ns": 458941406, "total_duration_ns": 2417691607, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.811, "decode_tok_s_wall_gross": 105.741, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 126.25, "median": 126.28, "max": 126.34, "mean": 126.29, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 126.29 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "power_w_all_cards": 126.29, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 250.4, "max_w": 317.77, "median_w": 258.79, "mean_w_whole_window": 209.1, "max_w_whole_window": 317.77, "limit_w": 350.0, "over_idle_w": 124.11, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 126.44, "median": 207.65, "max": 317.77, "mean": 209.1, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 61.3, "n": 9 }, "temperature_c": { "min": 54.0, "median": 59.0, "max": 61.0, "mean": 57.8, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1845.0, "max": 1875.0, "mean": 1811.7, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 146.41, "median": 258.79, "max": 317.77, "mean": 250.4, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 6 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 6 }, "temperature_c": { "min": 57.0, "median": 60.0, "max": 61.0, "mean": 59.7, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1815.0, "median": 1860.0, "max": 1875.0, "mean": 1855.0, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1833.58, "power_mean_w_all_cards": 250.4, "energy_j_per_1k_tokens_all_cards": 1833.58, "watts_per_tok_s": 1.8336, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 61.0, "temp_min_c": 57.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1815.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 250.4, "power_pct_of_limit": 71.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 61.0, "temp_min_c": 57.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1815.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 250.4, "power_pct_of_limit": 71.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 563.16, "first_token_was_thinking": false, "wall_s": 2.4312, "eval_count": 256, "eval_duration_ns": 1864988000, "decode_tok_s": 137.266, "prompt_eval_count": 536, "prompt_eval_duration_ns": 101437000, "prefill_tok_s": 5284.068, "load_duration_ns": 457564331, "total_duration_ns": 2427846640, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.509, "decode_tok_s_wall_gross": 105.299, "num_ctx_option": 49152, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 22.15, "median": 22.26, "max": 22.78, "mean": 22.4, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 22.4 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "power_w_all_cards": 22.4, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 4, "mean_w": 288.28, "max_w": 327.23, "median_w": 317.92, "mean_w_whole_window": 160.45, "max_w_whole_window": 327.23, "limit_w": 350.0, "over_idle_w": 265.88, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 22.19, "median": 153.0, "max": 327.23, "mean": 160.45, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 93.0, "mean": 41.3, "n": 9 }, "temperature_c": { "min": 51.0, "median": 59.0, "max": 61.0, "mean": 56.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1875.0, "max": 1875.0, "mean": 1486.7, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 190.04, "median": 317.92, "max": 327.23, "mean": 288.28, "n": 4 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 18879.0, "median": 18879.0, "max": 18879.0, "mean": 18879.0, "n": 4 }, "utilization_pct": { "min": 93.0, "median": 93.0, "max": 93.0, "mean": 93.0, "n": 4 }, "temperature_c": { "min": 60.0, "median": 60.5, "max": 61.0, "mean": 60.5, "n": 4 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 4 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 4 }, "clocks_sm_mhz": { "min": 1875.0, "median": 1875.0, "max": 1875.0, "mean": 1875.0, "n": 4 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 4 }, "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": 4 }, "busy_samples": 4 } } }, "energy_j_per_1k_tokens": 2100.15, "power_mean_w_all_cards": 288.28, "energy_j_per_1k_tokens_all_cards": 2100.15, "watts_per_tok_s": 2.1002, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 4, "temp_max_c": 61.0, "temp_min_c": 60.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 288.28, "power_pct_of_limit": 82.4, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 4, "temp_max_c": 61.0, "temp_min_c": 60.0, "temp_rise_c": 1.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1875.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 288.28, "power_pct_of_limit": 82.4, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 136.563, "median": 137.234, "max": 137.266, "mean": 137.021, "n": 3 }, "ttft_ms": { "min": 543.4, "median": 553.99, "max": 563.16, "mean": 553.52, "n": 3 }, "power_mean_w": { "min": 250.4, "median": 251.54, "max": 288.28, "mean": 263.41, "n": 3 }, "power_mean_w_all_cards": { "min": 250.4, "median": 251.54, "max": 288.28, "mean": 263.41, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1832.92, "median": 1833.58, "max": 2100.15, "mean": 1922.22, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1832.92, "median": 1833.58, "max": 2100.15, "mean": 1922.22, "n": 3 } }, "finished_utc": "2026-09-18T00:29:22Z" }, { "num_ctx": 65536, "started_utc": "2026-09-18T00:29:22Z", "loads": true, "size_total": 17490950224, "size_vram": 17490950224, "context_length_loaded": 65536, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 250.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 18879.0 }, "runner_peak_rss_gib": 1.35, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 11636, "rss_kb": 1412636, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3494 22315 104 21856 43545\nSwap: 24575 0 24575", "kv_footprint_bytes": 289931264, "kv_footprint_gib": 0.27, "runs": [ { "ttft_ms": 557.99, "first_token_was_thinking": false, "wall_s": 2.4245, "eval_count": 256, "eval_duration_ns": 1863475000, "decode_tok_s": 137.378, "prompt_eval_count": 536, "prompt_eval_duration_ns": 92092000, "prefill_tok_s": 5820.267, "load_duration_ns": 461679002, "total_duration_ns": 2421136672, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.619, "decode_tok_s_wall_gross": 105.589, "num_ctx_option": 65536, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 128.6, "median": 128.6, "max": 128.6, "mean": 128.6, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 128.6 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "power_w_all_cards": 128.6, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 256.47, "max_w": 325.89, "median_w": 268.28, "mean_w_whole_window": 213.88, "max_w_whole_window": 325.89, "limit_w": 350.0, "over_idle_w": 127.87, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 128.65, "median": 201.48, "max": 325.89, "mean": 213.88, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 61.3, "n": 9 }, "temperature_c": { "min": 56.0, "median": 61.0, "max": 63.0, "mean": 59.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1860.0, "max": 1875.0, "mean": 1818.3, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 151.29, "median": 268.28, "max": 325.89, "mean": 256.47, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 6 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 6 }, "temperature_c": { "min": 59.0, "median": 62.5, "max": 63.0, "mean": 61.8, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1830.0, "median": 1875.0, "max": 1875.0, "mean": 1865.0, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.6667, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1866.9, "power_mean_w_all_cards": 256.47, "energy_j_per_1k_tokens_all_cards": 1866.9, "watts_per_tok_s": 1.8669, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 63.0, "temp_min_c": 59.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 256.47, "power_pct_of_limit": 73.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 63.0, "temp_min_c": 59.0, "temp_rise_c": 4.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1875.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 256.47, "power_pct_of_limit": 73.3, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.6667, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 572.16, "first_token_was_thinking": false, "wall_s": 2.4477, "eval_count": 256, "eval_duration_ns": 1872613000, "decode_tok_s": 136.707, "prompt_eval_count": 536, "prompt_eval_duration_ns": 111438000, "prefill_tok_s": 4809.849, "load_duration_ns": 456215064, "total_duration_ns": 2444310657, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.958, "decode_tok_s_wall_gross": 104.586, "num_ctx_option": 65536, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 22.36, "median": 22.42, "max": 22.57, "mean": 22.45, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 22.45 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "power_w_all_cards": 22.45, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 254.84, "max_w": 319.12, "median_w": 292.12, "mean_w_whole_window": 155.45, "max_w_whole_window": 319.12, "limit_w": 350.0, "over_idle_w": 232.39, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 22.59, "median": 155.8, "max": 319.12, "mean": 155.45, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 51.1, "n": 9 }, "temperature_c": { "min": 53.0, "median": 60.0, "max": 62.0, "mean": 58.2, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1860.0, "max": 1860.0, "mean": 1435.0, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.4444, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 155.8, "median": 292.12, "max": 319.12, "mean": 254.84, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 5 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 5 }, "temperature_c": { "min": 60.0, "median": 62.0, "max": 62.0, "mean": 61.4, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1860.0, "max": 1860.0, "mean": 1860.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "throttle_sw_power_cap_fraction": 0.8, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1864.13, "power_mean_w_all_cards": 254.84, "energy_j_per_1k_tokens_all_cards": 1864.13, "watts_per_tok_s": 1.8641, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 62.0, "temp_min_c": 60.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 254.84, "power_pct_of_limit": 72.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 62.0, "temp_min_c": 60.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 254.84, "power_pct_of_limit": 72.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 573.63, "first_token_was_thinking": false, "wall_s": 2.4348, "eval_count": 256, "eval_duration_ns": 1858170000, "decode_tok_s": 137.77, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102762000, "prefill_tok_s": 5215.936, "load_duration_ns": 466741542, "total_duration_ns": 2431548390, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 137.01, "decode_tok_s_wall_gross": 105.142, "num_ctx_option": 65536, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 22.31, "median": 22.39, "max": 22.56, "mean": 22.42, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 22.42 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "power_w_all_cards": 22.42, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 271.9, "max_w": 338.18, "median_w": 324.0, "mean_w_whole_window": 165.62, "max_w_whole_window": 338.18, "limit_w": 350.0, "over_idle_w": 249.48, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 22.23, "median": 167.92, "max": 338.18, "mean": 165.62, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 51.1, "n": 9 }, "temperature_c": { "min": 51.0, "median": 60.0, "max": 62.0, "mean": 57.4, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1845.0, "max": 1905.0, "mean": 1475.0, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 167.92, "median": 324.0, "max": 338.18, "mean": 271.9, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19129.0, "median": 19129.0, "max": 19129.0, "mean": 19129.0, "n": 5 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 5 }, "temperature_c": { "min": 60.0, "median": 61.0, "max": 62.0, "mean": 61.0, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1830.0, "median": 1890.0, "max": 1905.0, "mean": 1875.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1973.58, "power_mean_w_all_cards": 271.9, "energy_j_per_1k_tokens_all_cards": 1973.58, "watts_per_tok_s": 1.9736, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 62.0, "temp_min_c": 60.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 271.9, "power_pct_of_limit": 77.7, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 62.0, "temp_min_c": 60.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1830.0, "clock_max_mhz": 1905.0, "clock_median_mhz": 1890.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 271.9, "power_pct_of_limit": 77.7, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 136.707, "median": 137.378, "max": 137.77, "mean": 137.285, "n": 3 }, "ttft_ms": { "min": 557.99, "median": 572.16, "max": 573.63, "mean": 567.93, "n": 3 }, "power_mean_w": { "min": 254.84, "median": 256.47, "max": 271.9, "mean": 261.07, "n": 3 }, "power_mean_w_all_cards": { "min": 254.84, "median": 256.47, "max": 271.9, "mean": 261.07, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1864.13, "median": 1866.9, "max": 1973.58, "mean": 1901.54, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1864.13, "median": 1866.9, "max": 1973.58, "mean": 1901.54, "n": 3 } }, "finished_utc": "2026-09-18T00:30:44Z" }, { "num_ctx": 98304, "started_utc": "2026-09-18T00:30:44Z", "loads": true, "size_total": 17658722384, "size_vram": 17658722384, "context_length_loaded": 98304, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 500.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19129.0 }, "runner_peak_rss_gib": 1.38, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 12535, "rss_kb": 1445912, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3588 22220 136 21889 43451\nSwap: 24575 0 24575", "kv_footprint_bytes": 457703424, "kv_footprint_gib": 0.43, "runs": [ { "ttft_ms": 553.88, "first_token_was_thinking": false, "wall_s": 2.4327, "eval_count": 256, "eval_duration_ns": 1875754000, "decode_tok_s": 136.478, "prompt_eval_count": 536, "prompt_eval_duration_ns": 92289000, "prefill_tok_s": 5807.843, "load_duration_ns": 457105256, "total_duration_ns": 2429324525, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.724, "decode_tok_s_wall_gross": 105.233, "num_ctx_option": 98304, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 128.96, "median": 129.0, "max": 129.09, "mean": 129.02, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 129.02 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "power_w_all_cards": 129.02, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 272.17, "max_w": 321.86, "median_w": 297.56, "mean_w_whole_window": 210.63, "max_w_whole_window": 321.86, "limit_w": 350.0, "over_idle_w": 143.15, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 129.17, "median": 202.72, "max": 321.86, "mean": 210.63, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 50.9, "n": 9 }, "temperature_c": { "min": 56.0, "median": 61.0, "max": 63.0, "mean": 60.1, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1860.0, "max": 1875.0, "mean": 1813.3, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 202.72, "median": 297.56, "max": 321.86, "mean": 272.17, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.6, "n": 5 }, "temperature_c": { "min": 61.0, "median": 63.0, "max": 63.0, "mean": 62.4, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1860.0, "max": 1875.0, "mean": 1866.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1994.23, "power_mean_w_all_cards": 272.17, "energy_j_per_1k_tokens_all_cards": 1994.23, "watts_per_tok_s": 1.9942, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 63.0, "temp_min_c": 61.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 272.17, "power_pct_of_limit": 77.8, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 63.0, "temp_min_c": 61.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 272.17, "power_pct_of_limit": 77.8, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 562.18, "first_token_was_thinking": false, "wall_s": 2.4373, "eval_count": 256, "eval_duration_ns": 1872052000, "decode_tok_s": 136.748, "prompt_eval_count": 536, "prompt_eval_duration_ns": 102007000, "prefill_tok_s": 5254.541, "load_duration_ns": 455856436, "total_duration_ns": 2433981622, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.99, "decode_tok_s_wall_gross": 105.033, "num_ctx_option": 98304, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 23.81, "median": 24.16, "max": 24.28, "mean": 24.08, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 24.08 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "power_w_all_cards": 24.08, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 257.5, "max_w": 322.14, "median_w": 306.56, "mean_w_whole_window": 158.34, "max_w_whole_window": 322.14, "limit_w": 350.0, "over_idle_w": 233.42, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 22.76, "median": 150.83, "max": 322.14, "mean": 158.34, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 50.9, "n": 9 }, "temperature_c": { "min": 53.0, "median": 61.0, "max": 63.0, "mean": 58.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1860.0, "max": 1875.0, "mean": 1480.0, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 150.83, "median": 306.56, "max": 322.14, "mean": 257.5, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.6, "n": 5 }, "temperature_c": { "min": 61.0, "median": 62.0, "max": 63.0, "mean": 62.2, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1860.0, "max": 1875.0, "mean": 1863.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1883.02, "power_mean_w_all_cards": 257.5, "energy_j_per_1k_tokens_all_cards": 1883.02, "watts_per_tok_s": 1.883, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 63.0, "temp_min_c": 61.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 257.5, "power_pct_of_limit": 73.6, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 63.0, "temp_min_c": 61.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1875.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 257.5, "power_pct_of_limit": 73.6, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 572.02, "first_token_was_thinking": false, "wall_s": 2.4488, "eval_count": 256, "eval_duration_ns": 1873810000, "decode_tok_s": 136.62, "prompt_eval_count": 536, "prompt_eval_duration_ns": 104104000, "prefill_tok_s": 5148.697, "load_duration_ns": 463711803, "total_duration_ns": 2445595707, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.868, "decode_tok_s_wall_gross": 104.539, "num_ctx_option": 98304, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 22.67, "median": 23.62, "max": 24.17, "mean": 23.49, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 23.49 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "power_w_all_cards": 23.49, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 257.94, "max_w": 319.41, "median_w": 297.51, "mean_w_whole_window": 157.37, "max_w_whole_window": 319.41, "limit_w": 350.0, "over_idle_w": 234.45, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 22.47, "median": 160.57, "max": 319.41, "mean": 157.37, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 51.1, "n": 9 }, "temperature_c": { "min": 53.0, "median": 61.0, "max": 63.0, "mean": 58.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1860.0, "max": 1860.0, "mean": 1460.0, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.7778, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 160.57, "median": 297.51, "max": 319.41, "mean": 257.94, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 19629.0, "median": 19629.0, "max": 19629.0, "mean": 19629.0, "n": 5 }, "utilization_pct": { "min": 92.0, "median": 92.0, "max": 92.0, "mean": 92.0, "n": 5 }, "temperature_c": { "min": 61.0, "median": 62.0, "max": 63.0, "mean": 62.2, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1860.0, "max": 1860.0, "mean": 1860.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 1888.01, "power_mean_w_all_cards": 257.94, "energy_j_per_1k_tokens_all_cards": 1888.01, "watts_per_tok_s": 1.888, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 63.0, "temp_min_c": 61.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 257.94, "power_pct_of_limit": 73.7, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 63.0, "temp_min_c": 61.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 257.94, "power_pct_of_limit": 73.7, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 136.478, "median": 136.62, "max": 136.748, "mean": 136.615, "n": 3 }, "ttft_ms": { "min": 553.88, "median": 562.18, "max": 572.02, "mean": 562.69, "n": 3 }, "power_mean_w": { "min": 257.5, "median": 257.94, "max": 272.17, "mean": 262.54, "n": 3 }, "power_mean_w_all_cards": { "min": 257.5, "median": 257.94, "max": 272.17, "mean": 262.54, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1883.02, "median": 1888.01, "max": 1994.23, "mean": 1921.75, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1883.02, "median": 1888.01, "max": 1994.23, "mean": 1921.75, "n": 3 } }, "finished_utc": "2026-09-18T00:31:55Z" }, { "num_ctx": 131072, "started_utc": "2026-09-18T00:31:55Z", "loads": true, "size_total": 17826494544, "size_vram": 17826494544, "context_length_loaded": 131072, "fits_fully_in_vram": true, "verdict": "fits: fully resident on the card", "offload_split": "100.0% VRAM / 0.0% RAM", "vram_delta_mib": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 500.0 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "memory_used_mib_per_card_before": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 19629.0 }, "runner_peak_rss_gib": 1.41, "runner_rss_within_allowance": true, "runner_procs": [ { "pid": 13380, "rss_kb": 1480112, "threads": 27 } ], "free_m_at_level": "total used free shared buff/cache available\nMem: 47040 3598 22208 168 21923 43442\nSwap: 24575 0 24575", "kv_footprint_bytes": 625475584, "kv_footprint_gib": 0.58, "runs": [ { "ttft_ms": 552.75, "first_token_was_thinking": false, "wall_s": 2.4297, "eval_count": 256, "eval_duration_ns": 1873841000, "decode_tok_s": 136.618, "prompt_eval_count": 536, "prompt_eval_duration_ns": 92342000, "prefill_tok_s": 5804.509, "load_duration_ns": 456007431, "total_duration_ns": 2426254746, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.86, "decode_tok_s_wall_gross": 105.363, "num_ctx_option": 131072, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 130.42, "median": 130.45, "max": 130.45, "mean": 130.44, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 130.44 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "power_w_all_cards": 130.44, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 5, "mean_w": 275.82, "max_w": 322.37, "median_w": 306.02, "mean_w_whole_window": 213.65, "max_w_whole_window": 322.37, "limit_w": 350.0, "over_idle_w": 145.38, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 130.43, "median": 202.18, "max": 322.37, "mean": 213.65, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 91.0, "max": 92.0, "mean": 50.9, "n": 9 }, "temperature_c": { "min": 58.0, "median": 63.0, "max": 65.0, "mean": 61.9, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1860.0, "max": 1860.0, "mean": 1813.3, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 202.18, "median": 306.02, "max": 322.37, "mean": 275.82, "n": 5 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 5 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.6, "n": 5 }, "temperature_c": { "min": 63.0, "median": 64.0, "max": 65.0, "mean": 64.2, "n": 5 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 5 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 5 }, "clocks_sm_mhz": { "min": 1860.0, "median": 1860.0, "max": 1860.0, "mean": 1860.0, "n": 5 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 5 }, "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": 5 }, "busy_samples": 5 } } }, "energy_j_per_1k_tokens": 2018.92, "power_mean_w_all_cards": 275.82, "energy_j_per_1k_tokens_all_cards": 2018.92, "watts_per_tok_s": 2.0189, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 65.0, "temp_min_c": 63.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 275.82, "power_pct_of_limit": 78.8, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 5, "temp_max_c": 65.0, "temp_min_c": 63.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 0.0, "fan_max_pct": 0.0, "clock_floor_mhz": 1860.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 275.82, "power_pct_of_limit": 78.8, "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": false, "verdict": "no clock drop during decode (the fans never spun up -- zero-RPM idle mode)" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 1 }, { "ttft_ms": 544.19, "first_token_was_thinking": false, "wall_s": 2.4187, "eval_count": 256, "eval_duration_ns": 1871518000, "decode_tok_s": 136.787, "prompt_eval_count": 536, "prompt_eval_duration_ns": 80975000, "prefill_tok_s": 6619.327, "load_duration_ns": 458682655, "total_duration_ns": 2415439512, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 136.033, "decode_tok_s_wall_gross": 105.84, "num_ctx_option": 131072, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 130.78, "median": 130.78, "max": 130.85, "mean": 130.8, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 130.8 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "power_w_all_cards": 130.8, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 7, "mean_w": 240.49, "max_w": 323.65, "median_w": 232.81, "mean_w_whole_window": 216.24, "max_w_whole_window": 323.65, "limit_w": 350.0, "over_idle_w": 109.69, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 131.36, "median": 216.78, "max": 323.65, "mean": 216.24, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 59.3, "n": 9 }, "temperature_c": { "min": 59.0, "median": 64.0, "max": 65.0, "mean": 62.4, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 43.0, "max": 53.0, "mean": 31.4, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1845.0, "max": 1860.0, "mean": 1811.7, "n": 9 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 131.75, "median": 232.81, "max": 323.65, "mean": 240.49, "n": 7 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 7 }, "utilization_pct": { "min": 37.0, "median": 92.0, "max": 92.0, "mean": 76.3, "n": 7 }, "temperature_c": { "min": 59.0, "median": 64.0, "max": 65.0, "mean": 63.4, "n": 7 }, "memory_temperature_c": null, "fan_pct": { "min": 0.0, "median": 43.0, "max": 48.0, "mean": 25.4, "n": 7 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 7 }, "clocks_sm_mhz": { "min": 1725.0, "median": 1860.0, "max": 1860.0, "mean": 1836.4, "n": 7 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 7 }, "throttle_sw_power_cap_fraction": 0.7143, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 7 }, "busy_samples": 7 } } }, "energy_j_per_1k_tokens": 1758.13, "power_mean_w_all_cards": 240.49, "energy_j_per_1k_tokens_all_cards": 1758.13, "watts_per_tok_s": 1.7581, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 65.0, "temp_min_c": 59.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 25.4, "fan_max_pct": 48.0, "clock_floor_mhz": 1725.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 350.0, "power_mean_w": 240.49, "power_pct_of_limit": 68.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.7143, "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 68.7% of the cap and the card at 65 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, "thermal_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 65.0, "temp_min_c": 59.0, "temp_rise_c": 6.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 16.0, "pcie_width_max": 16.0, "pcie_width_median": 16.0, "fan_mean_pct": 25.4, "fan_max_pct": 48.0, "clock_floor_mhz": 1725.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": true, "power_limit_w": 350.0, "power_mean_w": 240.49, "power_pct_of_limit": 68.7, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.7143, "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 68.7% of the cap and the card at 65 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 2 }, { "ttft_ms": 563.45, "first_token_was_thinking": false, "wall_s": 2.4423, "eval_count": 256, "eval_duration_ns": 1875832000, "decode_tok_s": 136.473, "prompt_eval_count": 536, "prompt_eval_duration_ns": 101545000, "prefill_tok_s": 5278.448, "load_duration_ns": 457604361, "total_duration_ns": 2439111326, "streamed_chunks": 256, "response_chars": 1023, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 135.718, "decode_tok_s_wall_gross": 104.817, "num_ctx_option": 131072, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 23.0, "median": 23.92, "max": 24.43, "mean": 23.78, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 6 }, "power_w_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 23.78 }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "power_w_all_cards": 23.78, "ups_whole_box": { "load_pct": null, "realpower_w": null, "realpower_nominal_w": null, "status": null, "note": "the UPS's reading of EVERYTHING it feeds -- the whole box, not the cards. The card figures beside it are board power from nvidia-smi. Neither is the other." } }, "measured_over": "samples where the GPU was busy", "busy_samples": 6, "mean_w": 222.39, "max_w": 316.96, "median_w": 244.14, "mean_w_whole_window": 156.68, "max_w_whole_window": 316.96, "limit_w": 350.0, "over_idle_w": 198.61, "samples": 9, "per_gpu": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 23.99, "median": 150.03, "max": 316.96, "mean": 156.68, "n": 9 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 9 }, "utilization_pct": { "min": 0.0, "median": 32.0, "max": 92.0, "mean": 41.3, "n": 9 }, "temperature_c": { "min": 53.0, "median": 61.0, "max": 63.0, "mean": 59.0, "n": 9 }, "memory_temperature_c": null, "fan_pct": { "min": 45.0, "median": 46.0, "max": 46.0, "mean": 45.6, "n": 9 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 }, "clocks_sm_mhz": { "min": 210.0, "median": 1845.0, "max": 1860.0, "mean": 1470.0, "n": 9 }, "clocks_mem_mhz": { "min": 405.0, "median": 9501.0, "max": 9501.0, "mean": 7479.7, "n": 9 }, "throttle_sw_power_cap_fraction": 0.5556, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 9, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 3090", "power_w": { "min": 67.25, "median": 244.14, "max": 316.96, "mean": 222.39, "n": 6 }, "power_limit_w": 350.0, "memory_used_mib": { "min": 20129.0, "median": 20129.0, "max": 20129.0, "mean": 20129.0, "n": 6 }, "utilization_pct": { "min": 32.0, "median": 62.0, "max": 92.0, "mean": 62.0, "n": 6 }, "temperature_c": { "min": 59.0, "median": 62.0, "max": 63.0, "mean": 61.5, "n": 6 }, "memory_temperature_c": null, "fan_pct": { "min": 45.0, "median": 45.0, "max": 46.0, "mean": 45.3, "n": 6 }, "pcie_width_current": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 6 }, "clocks_sm_mhz": { "min": 1800.0, "median": 1860.0, "max": 1860.0, "mean": 1847.5, "n": 6 }, "clocks_mem_mhz": { "min": 9501.0, "median": 9501.0, "max": 9501.0, "mean": 9501.0, "n": 6 }, "throttle_sw_power_cap_fraction": 0.8333, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 6 }, "busy_samples": 6 } } }, "energy_j_per_1k_tokens": 1629.56, "power_mean_w_all_cards": 222.39, "energy_j_per_1k_tokens_all_cards": 1629.56, "watts_per_tok_s": 1.6296, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 63.0, "temp_min_c": 59.0, "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": 45.3, "fan_max_pct": 46.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 222.39, "power_pct_of_limit": 63.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8333, "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-7aa0be10-974f-6430-52fe-09018a0e2e08": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 6, "temp_max_c": 63.0, "temp_min_c": 59.0, "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": 45.3, "fan_max_pct": 46.0, "clock_floor_mhz": 1800.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1860.0, "mem_clock_floor_mhz": 9501.0, "mem_clock_max_mhz": 9501.0, "mem_clock_held": true, "clock_dropped": false, "power_limit_w": 350.0, "power_mean_w": 222.39, "power_pct_of_limit": 63.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8333, "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" } }, "memory_used_mib_per_card": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": 20129.0 }, "pcie_width_under_load": { "GPU-7aa0be10-974f-6430-52fe-09018a0e2e08": { "min": 16.0, "median": 16.0, "max": 16.0, "mean": 16.0, "n": 9 } }, "ups_during_run": { "ups": "pr1500@localhost", "samples": 0, "load_pct": null, "whole_box_realpower_w": null, "whole_box_note": "the UPS's reading of everything it feeds, not the cards; nvidia-smi's per-card watts are a different measurement of a smaller thing", "stop_threshold_pct": 80.0, "breached": false }, "stop_conditions_hit": [], "run": 3 } ], "derived": { "decode_tok_s": { "min": 136.473, "median": 136.618, "max": 136.787, "mean": 136.626, "n": 3 }, "ttft_ms": { "min": 544.19, "median": 552.75, "max": 563.45, "mean": 553.46, "n": 3 }, "power_mean_w": { "min": 222.39, "median": 240.49, "max": 275.82, "mean": 246.23, "n": 3 }, "power_mean_w_all_cards": { "min": 222.39, "median": 240.49, "max": 275.82, "mean": 246.23, "n": 3 }, "energy_j_per_1k_tokens_all_cards": { "min": 1629.56, "median": 1758.13, "max": 2018.92, "mean": 1802.2, "n": 3 }, "energy_j_per_1k_tokens": { "min": 1629.56, "median": 1758.13, "max": 2018.92, "mean": 1802.2, "n": 3 } }, "finished_utc": "2026-09-18T00:32:57Z" } ], "errors": [], "context_length_max": 262144, "quantization": "Q4_K_M", "ups_after_arm": { "ups": "pr1500@localhost", "read_utc": "2026-09-18T00:32:57Z", "available": false, "error": "upsc rc=1: Error: Connection failure: Connection refused" }, "largest_context_that_fits": 131072, "levels_requested": [ 4096, 8192, 16384, 32768, 49152, 65536, 98304, 131072 ], "levels_not_reached": [], "ladder_topped_out": "every rung tested fits, up to and including 131,072 -- the ladder ran out of rungs before the card ran out of memory, so the figure reported is the largest window TESTED, not a measured ceiling", "headline": "the biggest context this arm holds for gemma4:26b is 131,072 tokens", "finished_utc": "2026-09-18T00:32:57Z" }