{ "model_tag": "gemma4:26b", "arm": "one-95w", "box_class": "gpu-5090-laptop-24g", "mode": "concurrency", "target_uuid": "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff", "cards": [ "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff" ], "card_labels": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": "soldered (mobile; no socket \u2014 link width is traced under load, never assumed) at 00000000:01:00.0 - vendor unread - GPU-edff232c" }, "card_identity": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": null, "bus_id": "00000000:01:00.0", "pci_sub_device_id": null, "vendor": null, "board_per_operator": null, "vbios": null, "serial": null, "pcie_link_width_max": null, "pcie_link_width_negotiated": null, "memory_total_mib": null, "memory_bus_width_bits": null, "clocks_max_sm_mhz": null, "clocks_max_memory_mhz": null, "power_default_limit_w": null, "power_min_limit_w": null, "power_max_limit_w": null, "power_limit_at_bench_start_w": null, "note": "the single NVIDIA GeForce RTX 5090 Laptop GPU 24 GB SOLDERED to this machine's board at 00000000:01:00.0. There is no seat, no partner and no card to swap. \u26a0 THE TRAP ON THIS BOX IS NOT A STALE BOOT UNIT, IT IS A RUNNING DAEMON: nvidia-powerd.service (NVIDIA Dynamic Boost) floats this board's power limit between its default and its maximum against the CPU's draw, continuously and without asking, and it has been running since 2026-09-08. The lesson every leg of this ladder carries \u2014 a cap read once at the start of a night is not a cap \u2014 is true here in its strongest form: such a reading is a sample of a moving signal. Every arm reads the limit back in the same call as its own data, every stage reads it again at its close, and a stage that finds it moved STOPS rather than relabelling its file. The second trap is a CLOCK LOCK: ai-perf.service runs `nvidia-smi -lgc 1200,2550` at boot on this box and the cards that produced the rows this bench compares against ran unlocked, so a floored clock changes what a power cap means and every record carries clocks.sm min and max so the confound is visible rather than implied." } }, "num_ctx": 4096, "instance_env": { "kind": "instance-env-receipt", "box_class": "gpu-5090-laptop-24g", "target_uuid": "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff", "written_utc": "2026-09-21T19:12:18Z", "unit": "bench-5090laptop-one", "shape": "one", "port": 11470, "base_url": "http://127.0.0.1:11470", "cuda_visible_devices": "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff", "OLLAMA_FLASH_ATTENTION": "1", "OLLAMA_KV_CACHE_TYPE": "q8_0", "OLLAMA_CONTEXT_LENGTH": 32768, "OLLAMA_": 4, "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" }, "ollama_binary": "/workshop/bench-laptop-5090-2026-09-21/ollama-0.32.13/bin/ollama", "ollama_client_version": "0.32.13", "ollama_version_pin": "0.32.13", "disclosed_tenants": "nomic-embed-text:latest on the SYSTEM ollama (:, OLLAMA_KEEP_ALIVE=-1, ~323 MB VRAM) \u2014 disclosed, never unloaded", "power_and_clocks_at_start_csv": "[N/A], 95.00 W, 95.00 W, 175.00 W, 1590 MHz, 9001 MHz, 3090 MHz", "power_and_clocks_at_start_fields": "power.limit,enforced.power.limit,power.default_limit,power.max_limit,clocks.sm,clocks.mem,clocks.max.sm", "nvidia_powerd_at_start": "inactive", "clock_lock_unit_at_start": "active", "clock_lock_unit_name": "ai-perf.service", "clock_lock_range_declared": "1200,2550", "clock_lock_unit_state_is_not_evidence": "ai-perf.service is Type=oneshot and reads 'active (exited)' for the whole uptime whatever happens to the card afterwards. The lock's real state is read from the CARD (clocks.sm against the lock's floor) and is recorded in this leg's clock-lock receipt.", "power_limits_at_start_w": "0, [N/A], 95.00 W;", "persistence_mode_at_start": "0, Enabled;", "pcie_link_width_at_start": "0, 8;" }, "num_gpu_option": null, "num_gpu_policy": "layers left to ollama's own planner (`auto`)", "num_gpu_note": "This arm MUST use the same layer policy as the arm it is compared against. On the pair of 3080s, left to the planner, this model spilled a quarter of itself to host RAM and the concurrency figures would have measured that spill rather than the cards; on one 24 GB card the planner is expected to place every layer, and the placement actually measured is recorded per run either way.", "memory_temp_support": { "checked_utc": "2026-09-21T19:12:18Z", "paths": { "nvidia-smi --query-gpu=temperature.memory": { "raw": "0, GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff, N/A", "readable": false }, "nvidia-smi -q -d TEMPERATURE": { "lines": [ "GPU Target Temperature : 87 C", "Memory Current Temp : N/A", "Memory Max Operating T.Limit 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." }, "ups_before_arm": { "ups": "pr1500@localhost", "read_utc": "2026-09-21T19:12:18Z", "available": false, "error": "no upsc on this box" }, "parallel_kv_note": "ollama allocates a KV cache PER PARALLEL SLOT, so a level of 4 streams at num_ctx N costs about four times the KV of one stream at N. The window this arm ran at is recorded above, and it is not automatically the largest window the card holds for a single stream.", "base_url": "http://127.0.0.1:11470", "started_utc": "2026-09-21T19:12:18Z", "prompt_sha256": "90eedd0c53f9554ae3837674504fcb7090013d9a432a653352183c0f25a7ce5c", "num_predict": 256, "trials_per_level": 3, "levels": [ { "level": 1, "contention_gate": { "attempts": [ { "window_s": 10.0, "samples": 20, "cards": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "utilization_pct": { "min": 0.0, "median": 0.0, "max": 94.0, "mean": 4.7, "n": 20 }, "power_w": { "min": 15.68, "median": 15.78, "max": 94.86, "mean": 22.23, "n": 20 }, "card_seen": true } }, "max_mean_util_pct": 4.7, "all_cards_seen": true, "attempt": 1, "verdict": "quiet: busiest card mean 4.7% <= 5.0%" } ], "passed": true, "cards": [ "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff" ] }, "trials": [ { "level": 1, "wall_s": 2.5014, "streams": [ { "ttft_ms": 345.52, "first_token_was_thinking": false, "wall_s": 2.5007, "eval_count": 256, "eval_duration_ns": 2152383000, "decode_tok_s": 118.938, "prompt_eval_count": 536, "prompt_eval_duration_ns": 73425000, "prefill_tok_s": 7299.966, "load_duration_ns": 266768872, "total_duration_ns": 2497029974, "streamed_chunks": 256, "response_chars": 1044, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 118.319, "decode_tok_s_wall_gross": 102.371, "num_ctx_option": 4096, "stream": 1 } ], "streams_ok": 1, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 118.938 ], "per_stream_ttft_ms": [ 345.52 ], "aggregate_sum_tok_s": 118.938, "aggregate_wall_tok_s": 102.341, "tokens_decoded_total": 256, "per_stream_decode": { "min": 118.938, "median": 118.938, "max": 118.938, "mean": 118.938, "n": 1 }, "per_stream_ttft": { "min": 345.52, "median": 345.52, "max": 345.52, "mean": 345.52, "n": 1 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 15.66, "median": 15.7, "max": 15.76, "mean": 15.71, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 15.71 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 15.71, "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." } }, "mean_w": 85.7, "max_w": 95.09, "over_idle_w": 69.99, "samples": 10, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.73, "median": 82.56, "max": 95.09, "mean": 66.47, "n": 10 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 10 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 93.0, "max": 96.0, "mean": 65.9, "n": 10 }, "temperature_c": { "min": 36.0, "median": 41.0, "max": 42.0, "mean": 40.2, "n": 10 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 10 }, "clocks_sm_mhz": { "min": 1027.0, "median": 1665.0, "max": 2077.0, "mean": 1597.2, "n": 10 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 13001.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.7, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 58.01, "median": 93.78, "max": 95.09, "mean": 85.7, "n": 7 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 7 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 7 }, "utilization_pct": { "min": 93.0, "median": 93.0, "max": 96.0, "mean": 94.1, "n": 7 }, "temperature_c": { "min": 40.0, "median": 41.0, "max": 42.0, "mean": 41.1, "n": 7 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 7 }, "clocks_sm_mhz": { "min": 1027.0, "median": 1740.0, "max": 1785.0, "mean": 1530.7, "n": 7 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "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": 837.4, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 42.0, "temp_min_c": 40.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1027.0, "clock_max_mhz": 1785.0, "clock_median_mhz": 1740.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 85.7, "power_pct_of_limit": 90.2, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 90.2% of the 95 W cap, at 42 C -- well below any throttling temperature" } }, { "level": 1, "wall_s": 2.5042, "streams": [ { "ttft_ms": 337.99, "first_token_was_thinking": false, "wall_s": 2.5038, "eval_count": 256, "eval_duration_ns": 2162746000, "decode_tok_s": 118.368, "prompt_eval_count": 536, "prompt_eval_duration_ns": 73181000, "prefill_tok_s": 7324.305, "load_duration_ns": 259556951, "total_duration_ns": 2500227249, "streamed_chunks": 256, "response_chars": 1044, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 117.738, "decode_tok_s_wall_gross": 102.244, "num_ctx_option": 4096, "stream": 1 } ], "streams_ok": 1, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 118.368 ], "per_stream_ttft_ms": [ 337.99 ], "aggregate_sum_tok_s": 118.368, "aggregate_wall_tok_s": 102.226, "tokens_decoded_total": 256, "per_stream_decode": { "min": 118.368, "median": 118.368, "max": 118.368, "mean": 118.368, "n": 1 }, "per_stream_ttft": { "min": 337.99, "median": 337.99, "max": 337.99, "mean": 337.99, "n": 1 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.15, "median": 19.45, "max": 93.23, "mean": 37.09, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 95.0, "mean": 15.8, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 37.09 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 37.09, "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." } }, "mean_w": 88.26, "max_w": 97.92, "over_idle_w": 51.17, "samples": 10, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.79, "median": 89.63, "max": 97.92, "mean": 68.19, "n": 10 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 10 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 93.0, "max": 96.0, "mean": 66.0, "n": 10 }, "temperature_c": { "min": 37.0, "median": 41.0, "max": 42.0, "mean": 40.4, "n": 10 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 10 }, "clocks_sm_mhz": { "min": 1050.0, "median": 1665.0, "max": 2160.0, "mean": 1614.6, "n": 10 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 13001.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.7, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 57.77, "median": 93.36, "max": 97.92, "mean": 88.26, "n": 7 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 7 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 7 }, "utilization_pct": { "min": 93.0, "median": 93.0, "max": 96.0, "mean": 94.3, "n": 7 }, "temperature_c": { "min": 40.0, "median": 41.0, "max": 42.0, "mean": 41.3, "n": 7 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 7 }, "clocks_sm_mhz": { "min": 1050.0, "median": 1740.0, "max": 1852.0, "mean": 1543.7, "n": 7 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "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": 863.38, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 42.0, "temp_min_c": 40.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1050.0, "clock_max_mhz": 1852.0, "clock_median_mhz": 1740.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 88.26, "power_pct_of_limit": 92.9, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 92.9% of the 95 W cap, at 42 C -- well below any throttling temperature" } }, { "level": 1, "wall_s": 2.5438, "streams": [ { "ttft_ms": 376.03, "first_token_was_thinking": false, "wall_s": 2.5432, "eval_count": 256, "eval_duration_ns": 2164182000, "decode_tok_s": 118.289, "prompt_eval_count": 536, "prompt_eval_duration_ns": 70146000, "prefill_tok_s": 7641.205, "load_duration_ns": 300874109, "total_duration_ns": 2539571246, "streamed_chunks": 256, "response_chars": 1044, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 117.665, "decode_tok_s_wall_gross": 100.66, "num_ctx_option": 4096, "stream": 1 } ], "streams_ok": 1, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 118.289 ], "per_stream_ttft_ms": [ 376.03 ], "aggregate_sum_tok_s": 118.289, "aggregate_wall_tok_s": 100.636, "tokens_decoded_total": 256, "per_stream_decode": { "min": 118.289, "median": 118.289, "max": 118.289, "mean": 118.289, "n": 1 }, "per_stream_ttft": { "min": 376.03, "median": 376.03, "max": 376.03, "mean": 376.03, "n": 1 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.32, "median": 23.71, "max": 94.61, "mean": 38.91, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 96.0, "mean": 16.0, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 38.91 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 38.91, "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." } }, "mean_w": 86.32, "max_w": 94.94, "over_idle_w": 47.41, "samples": 10, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.85, "median": 86.05, "max": 94.94, "mean": 66.49, "n": 10 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 10 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 10 }, "utilization_pct": { "min": 0.0, "median": 93.5, "max": 95.0, "mean": 65.9, "n": 10 }, "temperature_c": { "min": 37.0, "median": 41.5, "max": 43.0, "mean": 40.7, "n": 10 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 10 }, "clocks_sm_mhz": { "min": 1222.0, "median": 1680.0, "max": 2032.0, "mean": 1631.8, "n": 10 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 12501.0, "n": 10 }, "throttle_sw_power_cap_fraction": 0.7, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 10, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 54.36, "median": 93.68, "max": 94.94, "mean": 86.32, "n": 7 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 7 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 7 }, "utilization_pct": { "min": 93.0, "median": 94.0, "max": 95.0, "mean": 94.1, "n": 7 }, "temperature_c": { "min": 40.0, "median": 42.0, "max": 43.0, "mean": 41.7, "n": 7 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 7 }, "clocks_sm_mhz": { "min": 1222.0, "median": 1770.0, "max": 2032.0, "mean": 1587.6, "n": 7 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "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": 857.75, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 43.0, "temp_min_c": 40.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1222.0, "clock_max_mhz": 2032.0, "clock_median_mhz": 1770.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 86.32, "power_pct_of_limit": 90.9, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": true, "verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 90.9% of the 95 W cap, at 43 C -- well below any throttling temperature" } } ], "spread_gate": { "limit": 0.15, "values": [ 118.938, 118.368, 118.289 ], "spread_fraction": 0.0055, "verdict": "spread 0.5% of the median within the 15% gate" }, "voided_by_gate": false, "derived": { "aggregate_sum_tok_s": { "min": 118.289, "median": 118.368, "max": 118.938, "mean": 118.532, "n": 3 }, "aggregate_wall_tok_s": { "min": 100.636, "median": 102.226, "max": 102.341, "mean": 101.734, "n": 3 }, "per_stream_decode_tok_s": { "min": 118.289, "median": 118.368, "max": 118.938, "mean": 118.532, "n": 3 }, "per_stream_ttft_ms": { "min": 337.99, "median": 345.52, "max": 376.03, "mean": 353.18, "n": 3 }, "power_mean_w": { "min": 85.7, "median": 86.32, "max": 88.26, "mean": 86.76, "n": 3 }, "energy_j_per_1k_tokens": { "min": 837.4, "median": 857.75, "max": 863.38, "mean": 852.84, "n": 3 }, "temp_max_c": { "min": 42.0, "median": 42.0, "max": 43.0, "mean": 42.3, "n": 3 }, "fan_mean_pct": null, "clock_floor_mhz": { "min": 1027.0, "median": 1050.0, "max": 1222.0, "mean": 1099.7, "n": 3 } }, "thermal_verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 90.2% of the 95 W cap, at 42 C -- well below any throttling temperature; the POWER CAP doing its job: the SM clock fell while draw sat at 92.9% of the 95 W cap, at 42 C -- well below any throttling temperature; the POWER CAP doing its job: the SM clock fell while draw sat at 90.9% of the 95 W cap, at 43 C -- well below any throttling temperature" }, { "level": 2, "contention_gate": { "attempts": [ { "window_s": 10.0, "samples": 20, "cards": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "utilization_pct": { "min": 0.0, "median": 0.0, "max": 95.0, "mean": 4.8, "n": 20 }, "power_w": { "min": 15.78, "median": 15.85, "max": 94.94, "mean": 22.25, "n": 20 }, "card_seen": true } }, "max_mean_util_pct": 4.8, "all_cards_seen": true, "attempt": 1, "verdict": "quiet: busiest card mean 4.8% <= 5.0%" } ], "passed": true, "cards": [ "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff" ] }, "trials": [ { "level": 2, "wall_s": 4.5132, "streams": [ { "ttft_ms": 1990.16, "first_token_was_thinking": false, "wall_s": 4.5121, "eval_count": 256, "eval_duration_ns": 2518921000, "decode_tok_s": 101.631, "prompt_eval_count": 536, "prompt_eval_duration_ns": 1625131000, "prefill_tok_s": 329.82, "load_duration_ns": 321767201, "total_duration_ns": 4508743273, "streamed_chunks": 256, "response_chars": 1041, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 101.112, "decode_tok_s_wall_gross": 56.736, "num_ctx_option": 4096, "stream": 1 }, { "ttft_ms": 1829.53, "first_token_was_thinking": false, "wall_s": 4.4898, "eval_count": 256, "eval_duration_ns": 2656298000, "decode_tok_s": 96.375, "prompt_eval_count": 536, "prompt_eval_duration_ns": 1504154000, "prefill_tok_s": 356.346, "load_duration_ns": 320848869, "total_duration_ns": 4485492820, "streamed_chunks": 256, "response_chars": 1026, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 95.856, "decode_tok_s_wall_gross": 57.019, "num_ctx_option": 4096, "stream": 2 } ], "streams_ok": 2, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 101.631, 96.375 ], "per_stream_ttft_ms": [ 1990.16, 1829.53 ], "aggregate_sum_tok_s": 198.006, "aggregate_wall_tok_s": 113.446, "tokens_decoded_total": 512, "per_stream_decode": { "min": 96.375, "median": 99.003, "max": 101.631, "mean": 99.003, "n": 2 }, "per_stream_ttft": { "min": 1829.53, "median": 1909.85, "max": 1990.16, "mean": 1909.85, "n": 2 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 15.79, "median": 15.82, "max": 15.85, "mean": 15.82, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 15.82 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 15.82, "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." } }, "mean_w": 94.6, "max_w": 96.73, "over_idle_w": 78.78, "samples": 17, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.87, "median": 47.97, "max": 96.73, "mean": 54.16, "n": 17 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 17 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 17 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 94.0, "mean": 38.0, "n": 17 }, "temperature_c": { "min": 37.0, "median": 41.0, "max": 42.0, "mean": 39.6, "n": 17 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 17 }, "clocks_sm_mhz": { "min": 1050.0, "median": 1590.0, "max": 1807.0, "mean": 1552.4, "n": 17 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 11942.2, "n": 17 }, "throttle_sw_power_cap_fraction": 0.5294, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 17, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 93.12, "median": 94.2, "max": 96.73, "mean": 94.6, "n": 7 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 7 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 7 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 94.0, "mean": 92.3, "n": 7 }, "temperature_c": { "min": 40.0, "median": 42.0, "max": 42.0, "mean": 41.4, "n": 7 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 7 }, "clocks_sm_mhz": { "min": 1050.0, "median": 1432.0, "max": 1702.0, "mean": 1444.0, "n": 7 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "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": 833.88, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 7, "temp_max_c": 42.0, "temp_min_c": 40.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1050.0, "clock_max_mhz": 1702.0, "clock_median_mhz": 1432.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 94.6, "power_pct_of_limit": 99.6, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 99.6% of the 95 W cap, at 42 C -- well below any throttling temperature" } }, { "level": 2, "wall_s": 3.0598, "streams": [ { "ttft_ms": 500.89, "first_token_was_thinking": false, "wall_s": 3.0589, "eval_count": 256, "eval_duration_ns": 2555172000, "decode_tok_s": 100.189, "prompt_eval_count": 536, "prompt_eval_duration_ns": 123388000, "prefill_tok_s": 4344.02, "load_duration_ns": 301426553, "total_duration_ns": 3055424092, "streamed_chunks": 256, "response_chars": 1040, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 99.685, "decode_tok_s_wall_gross": 83.689, "num_ctx_option": 4096, "stream": 1 }, { "ttft_ms": 501.27, "first_token_was_thinking": false, "wall_s": 3.0596, "eval_count": 256, "eval_duration_ns": 2555110000, "decode_tok_s": 100.191, "prompt_eval_count": 536, "prompt_eval_duration_ns": 95554000, "prefill_tok_s": 5609.394, "load_duration_ns": 323588606, "total_duration_ns": 3056142873, "streamed_chunks": 256, "response_chars": 1031, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 99.675, "decode_tok_s_wall_gross": 83.672, "num_ctx_option": 4096, "stream": 2 } ], "streams_ok": 2, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 100.189, 100.191 ], "per_stream_ttft_ms": [ 500.89, 501.27 ], "aggregate_sum_tok_s": 200.38, "aggregate_wall_tok_s": 167.331, "tokens_decoded_total": 512, "per_stream_decode": { "min": 100.189, "median": 100.19, "max": 100.191, "mean": 100.19, "n": 2 }, "per_stream_ttft": { "min": 500.89, "median": 501.08, "max": 501.27, "mean": 501.08, "n": 2 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 15.95, "median": 19.76, "max": 93.32, "mean": 37.32, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 92.0, "mean": 15.3, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 37.32 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 37.32, "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." } }, "mean_w": 84.38, "max_w": 95.65, "over_idle_w": 47.06, "samples": 12, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.92, "median": 87.65, "max": 95.65, "mean": 67.8, "n": 12 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 12 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 12 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 68.8, "n": 12 }, "temperature_c": { "min": 38.0, "median": 42.0, "max": 43.0, "mean": 41.4, "n": 12 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 12 }, "clocks_sm_mhz": { "min": 915.0, "median": 1638.5, "max": 1860.0, "mean": 1542.2, "n": 12 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 12751.0, "n": 12 }, "throttle_sw_power_cap_fraction": 0.75, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 12, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 49.54, "median": 92.8, "max": 95.65, "mean": 84.38, "n": 9 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 9 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 9 }, "utilization_pct": { "min": 91.0, "median": 92.0, "max": 92.0, "mean": 91.8, "n": 9 }, "temperature_c": { "min": 41.0, "median": 42.0, "max": 43.0, "mean": 42.2, "n": 9 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 9 }, "clocks_sm_mhz": { "min": 915.0, "median": 1657.0, "max": 1860.0, "mean": 1522.2, "n": 9 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "n": 9 }, "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": 9 }, "busy_samples": 9 } } }, "energy_j_per_1k_tokens": 504.27, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 9, "temp_max_c": 43.0, "temp_min_c": 41.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 915.0, "clock_max_mhz": 1860.0, "clock_median_mhz": 1657.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 84.38, "power_pct_of_limit": 88.8, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "NEITHER cap: the SM clock varied with draw at 88.8% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" } }, { "level": 2, "wall_s": 3.097, "streams": [ { "ttft_ms": 525.67, "first_token_was_thinking": false, "wall_s": 3.096, "eval_count": 256, "eval_duration_ns": 2567632000, "decode_tok_s": 99.703, "prompt_eval_count": 536, "prompt_eval_duration_ns": 125869000, "prefill_tok_s": 4258.396, "load_duration_ns": 322754714, "total_duration_ns": 3092775130, "streamed_chunks": 256, "response_chars": 1040, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 99.211, "decode_tok_s_wall_gross": 82.688, "num_ctx_option": 4096, "stream": 1 }, { "ttft_ms": 526.19, "first_token_was_thinking": false, "wall_s": 3.0966, "eval_count": 256, "eval_duration_ns": 2567435000, "decode_tok_s": 99.71, "prompt_eval_count": 536, "prompt_eval_duration_ns": 95117000, "prefill_tok_s": 5635.165, "load_duration_ns": 337872961, "total_duration_ns": 3093645408, "streamed_chunks": 256, "response_chars": 1031, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 99.204, "decode_tok_s_wall_gross": 82.67, "num_ctx_option": 4096, "stream": 2 } ], "streams_ok": 2, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 99.703, 99.71 ], "per_stream_ttft_ms": [ 525.67, 526.19 ], "aggregate_sum_tok_s": 199.413, "aggregate_wall_tok_s": 165.321, "tokens_decoded_total": 512, "per_stream_decode": { "min": 99.703, "median": 99.707, "max": 99.71, "mean": 99.707, "n": 2 }, "per_stream_ttft": { "min": 525.67, "median": 525.93, "max": 526.19, "mean": 525.93, "n": 2 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 16.23, "median": 20.36, "max": 93.83, "mean": 38.99, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 92.0, "max": 92.0, "mean": 61.3, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 38.99 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 38.99, "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." } }, "mean_w": 83.14, "max_w": 95.08, "over_idle_w": 44.15, "samples": 12, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.87, "median": 87.68, "max": 95.08, "mean": 66.89, "n": 12 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 12 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 12 }, "utilization_pct": { "min": 0.0, "median": 74.5, "max": 94.0, "mean": 60.8, "n": 12 }, "temperature_c": { "min": 38.0, "median": 42.0, "max": 43.0, "mean": 41.6, "n": 12 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 12 }, "clocks_sm_mhz": { "min": 1005.0, "median": 1590.0, "max": 1852.0, "mean": 1528.5, "n": 12 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 12751.0, "n": 12 }, "throttle_sw_power_cap_fraction": 0.75, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 12, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 38.39, "median": 93.77, "max": 95.08, "mean": 83.14, "n": 9 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 9 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 9 }, "utilization_pct": { "min": 57.0, "median": 92.0, "max": 94.0, "mean": 81.0, "n": 9 }, "temperature_c": { "min": 41.0, "median": 43.0, "max": 43.0, "mean": 42.4, "n": 9 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 9 }, "clocks_sm_mhz": { "min": 1005.0, "median": 1642.0, "max": 1852.0, "mean": 1508.0, "n": 9 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "n": 9 }, "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": 9 }, "busy_samples": 9 } } }, "energy_j_per_1k_tokens": 502.9, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 9, "temp_max_c": 43.0, "temp_min_c": 41.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1005.0, "clock_max_mhz": 1852.0, "clock_median_mhz": 1642.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 83.14, "power_pct_of_limit": 87.5, "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": false, "verdict": "NEITHER cap: the SM clock varied with draw at 87.5% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" } } ], "spread_gate": { "limit": 0.15, "values": [ 198.006, 200.38, 199.413 ], "spread_fraction": 0.0119, "verdict": "spread 1.2% of the median within the 15% gate" }, "voided_by_gate": false, "derived": { "aggregate_sum_tok_s": { "min": 198.006, "median": 199.413, "max": 200.38, "mean": 199.266, "n": 3 }, "aggregate_wall_tok_s": { "min": 113.446, "median": 165.321, "max": 167.331, "mean": 148.699, "n": 3 }, "per_stream_decode_tok_s": { "min": 96.375, "median": 99.95, "max": 101.631, "mean": 99.633, "n": 6 }, "per_stream_ttft_ms": { "min": 500.89, "median": 525.93, "max": 1990.16, "mean": 978.95, "n": 6 }, "power_mean_w": { "min": 83.14, "median": 84.38, "max": 94.6, "mean": 87.37, "n": 3 }, "energy_j_per_1k_tokens": { "min": 502.9, "median": 504.27, "max": 833.88, "mean": 613.68, "n": 3 }, "temp_max_c": { "min": 42.0, "median": 43.0, "max": 43.0, "mean": 42.7, "n": 3 }, "fan_mean_pct": null, "clock_floor_mhz": { "min": 915.0, "median": 1005.0, "max": 1050.0, "mean": 990.0, "n": 3 } }, "thermal_verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 99.6% of the 95 W cap, at 42 C -- well below any throttling temperature; NEITHER cap: the SM clock varied with draw at 88.8% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it; NEITHER cap: the SM clock varied with draw at 87.5% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" }, { "level": 4, "contention_gate": { "attempts": [ { "window_s": 10.0, "samples": 20, "cards": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "utilization_pct": { "min": 0.0, "median": 0.0, "max": 93.0, "mean": 37.2, "n": 20 }, "power_w": { "min": 15.8, "median": 16.2, "max": 93.77, "mean": 22.97, "n": 20 }, "card_seen": true } }, "max_mean_util_pct": 37.2, "all_cards_seen": true, "attempt": 1, "verdict": "busy: busiest card mean 37.2% > 5.0% -- waiting" }, { "window_s": 10.0, "samples": 20, "cards": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 20 }, "power_w": { "min": 6.57, "median": 15.86, "max": 16.19, "mean": 14.11, "n": 20 }, "card_seen": true } }, "max_mean_util_pct": 0.0, "all_cards_seen": true, "attempt": 2, "verdict": "quiet: busiest card mean 0.0% <= 5.0%" } ], "passed": true, "cards": [ "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff" ] }, "trials": [ { "level": 4, "wall_s": 4.349, "streams": [ { "ttft_ms": 652.98, "first_token_was_thinking": false, "wall_s": 4.348, "eval_count": 256, "eval_duration_ns": 3692812000, "decode_tok_s": 69.324, "prompt_eval_count": 536, "prompt_eval_duration_ns": 159389000, "prefill_tok_s": 3362.842, "load_duration_ns": 326879120, "total_duration_ns": 4345181869, "streamed_chunks": 256, "response_chars": 1043, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 69.012, "decode_tok_s_wall_gross": 58.878, "num_ctx_option": 4096, "stream": 1 }, { "ttft_ms": 651.89, "first_token_was_thinking": false, "wall_s": 4.3474, "eval_count": 256, "eval_duration_ns": 3693279000, "decode_tok_s": 69.315, "prompt_eval_count": 536, "prompt_eval_duration_ns": 210931000, "prefill_tok_s": 2541.115, "load_duration_ns": 335929791, "total_duration_ns": 4344775836, "streamed_chunks": 256, "response_chars": 1051, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 69.003, "decode_tok_s_wall_gross": 58.886, "num_ctx_option": 4096, "stream": 2 }, { "ttft_ms": 547.32, "first_token_was_thinking": false, "wall_s": 4.3271, "eval_count": 256, "eval_duration_ns": 3776689000, "decode_tok_s": 67.784, "prompt_eval_count": 536, "prompt_eval_duration_ns": 217368000, "prefill_tok_s": 2465.864, "load_duration_ns": 324822985, "total_duration_ns": 4323611822, "streamed_chunks": 256, "response_chars": 1032, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 67.464, "decode_tok_s_wall_gross": 59.162, "num_ctx_option": 4096, "stream": 3 }, { "ttft_ms": 652.26, "first_token_was_thinking": false, "wall_s": 4.3481, "eval_count": 256, "eval_duration_ns": 3693582000, "decode_tok_s": 69.309, "prompt_eval_count": 536, "prompt_eval_duration_ns": 240536000, "prefill_tok_s": 2228.357, "load_duration_ns": 359028966, "total_duration_ns": 4344515720, "streamed_chunks": 256, "response_chars": 1051, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.997, "decode_tok_s_wall_gross": 58.876, "num_ctx_option": 4096, "stream": 4 } ], "streams_ok": 4, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 69.324, 69.315, 67.784, 69.309 ], "per_stream_ttft_ms": [ 652.98, 651.89, 547.32, 652.26 ], "aggregate_sum_tok_s": 275.732, "aggregate_wall_tok_s": 235.458, "tokens_decoded_total": 1024, "per_stream_decode": { "min": 67.784, "median": 69.312, "max": 69.324, "mean": 68.933, "n": 4 }, "per_stream_ttft": { "min": 547.32, "median": 652.08, "max": 652.98, "mean": 626.11, "n": 4 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 6.57, "median": 6.62, "max": 6.65, "mean": 6.62, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 0.0, "mean": 0.0, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 6.62 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 6.62, "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." } }, "mean_w": 78.34, "max_w": 94.33, "over_idle_w": 71.72, "samples": 16, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 6.62, "median": 92.78, "max": 94.33, "mean": 69.38, "n": 16 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 16 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 16 }, "utilization_pct": { "min": 0.0, "median": 89.0, "max": 92.0, "mean": 70.3, "n": 16 }, "temperature_c": { "min": 37.0, "median": 42.0, "max": 43.0, "mean": 41.0, "n": 16 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 16 }, "clocks_sm_mhz": { "min": 180.0, "median": 1447.0, "max": 1777.0, "mean": 1303.8, "n": 16 }, "clocks_mem_mhz": { "min": 405.0, "median": 14001.0, "max": 14001.0, "mean": 11989.0, "n": 16 }, "throttle_sw_power_cap_fraction": 0.75, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 16, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 11.87, "median": 93.77, "max": 94.33, "mean": 78.34, "n": 14 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 14 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 14 }, "utilization_pct": { "min": 23.0, "median": 89.5, "max": 92.0, "mean": 80.4, "n": 14 }, "temperature_c": { "min": 40.0, "median": 42.0, "max": 43.0, "mean": 41.6, "n": 14 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 14 }, "clocks_sm_mhz": { "min": 1230.0, "median": 1511.0, "max": 1777.0, "mean": 1464.4, "n": 14 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 13643.9, "n": 14 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 14 }, "busy_samples": 14 } } }, "energy_j_per_1k_tokens": 332.71, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 14, "temp_max_c": 43.0, "temp_min_c": 40.0, "temp_rise_c": 3.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1230.0, "clock_max_mhz": 1777.0, "clock_median_mhz": 1511.0, "mem_clock_floor_mhz": 9001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": false, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 78.34, "power_pct_of_limit": 82.5, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8571, "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 82.5% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" } }, { "level": 4, "wall_s": 6.3376, "streams": [ { "ttft_ms": 2637.92, "first_token_was_thinking": false, "wall_s": 6.3366, "eval_count": 256, "eval_duration_ns": 3696389000, "decode_tok_s": 69.257, "prompt_eval_count": 536, "prompt_eval_duration_ns": 2004872000, "prefill_tok_s": 267.349, "load_duration_ns": 365274137, "total_duration_ns": 6334089420, "streamed_chunks": 256, "response_chars": 1041, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.944, "decode_tok_s_wall_gross": 40.4, "num_ctx_option": 4096, "stream": 1 }, { "ttft_ms": 2637.61, "first_token_was_thinking": false, "wall_s": 6.3367, "eval_count": 256, "eval_duration_ns": 3696466000, "decode_tok_s": 69.255, "prompt_eval_count": 536, "prompt_eval_duration_ns": 2068206999, "prefill_tok_s": 259.162, "load_duration_ns": 318077651, "total_duration_ns": 6333141445, "streamed_chunks": 256, "response_chars": 1041, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.936, "decode_tok_s_wall_gross": 40.4, "num_ctx_option": 4096, "stream": 2 }, { "ttft_ms": 2637.62, "first_token_was_thinking": false, "wall_s": 6.337, "eval_count": 256, "eval_duration_ns": 3696501000, "decode_tok_s": 69.255, "prompt_eval_count": 536, "prompt_eval_duration_ns": 2099757999, "prefill_tok_s": 255.268, "load_duration_ns": 316263187, "total_duration_ns": 6333230197, "streamed_chunks": 256, "response_chars": 1043, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.931, "decode_tok_s_wall_gross": 40.398, "num_ctx_option": 4096, "stream": 3 }, { "ttft_ms": 2638.21, "first_token_was_thinking": false, "wall_s": 6.3366, "eval_count": 256, "eval_duration_ns": 3696415000, "decode_tok_s": 69.256, "prompt_eval_count": 536, "prompt_eval_duration_ns": 2037132000, "prefill_tok_s": 263.115, "load_duration_ns": 335465132, "total_duration_ns": 6333757662, "streamed_chunks": 256, "response_chars": 1044, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.949, "decode_tok_s_wall_gross": 40.4, "num_ctx_option": 4096, "stream": 4 } ], "streams_ok": 4, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 69.257, 69.255, 69.255, 69.256 ], "per_stream_ttft_ms": [ 2637.92, 2637.61, 2637.62, 2638.21 ], "aggregate_sum_tok_s": 277.023, "aggregate_wall_tok_s": 161.574, "tokens_decoded_total": 1024, "per_stream_decode": { "min": 69.255, "median": 69.255, "max": 69.257, "mean": 69.256, "n": 4 }, "per_stream_ttft": { "min": 2637.61, "median": 2637.77, "max": 2638.21, "mean": 2637.84, "n": 4 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 15.97, "median": 20.41, "max": 92.83, "mean": 38.21, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 90.0, "mean": 15.0, "n": 6 }, "memory_used_mib": { "min": 18561.0, "median": 18561.0, "max": 18561.0, "mean": 18561.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 38.21 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18561.0 }, "power_w_all_cards": 38.21, "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." } }, "mean_w": 80.75, "max_w": 94.31, "over_idle_w": 42.54, "samples": 24, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.89, "median": 58.59, "max": 94.31, "mean": 55.5, "n": 24 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 24 }, "memory_used_mib": { "min": 18561.0, "median": 18563.0, "max": 18563.0, "mean": 18562.8, "n": 24 }, "utilization_pct": { "min": 0.0, "median": 51.5, "max": 90.0, "mean": 45.4, "n": 24 }, "temperature_c": { "min": 38.0, "median": 42.0, "max": 43.0, "mean": 40.7, "n": 24 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 24 }, "clocks_sm_mhz": { "min": 1020.0, "median": 1590.0, "max": 1837.0, "mean": 1538.3, "n": 24 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 11709.3, "n": 24 }, "throttle_sw_power_cap_fraction": 0.5, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 24, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 17.72, "median": 93.47, "max": 94.31, "mean": 80.75, "n": 14 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 14 }, "memory_used_mib": { "min": 18563.0, "median": 18563.0, "max": 18563.0, "mean": 18563.0, "n": 14 }, "utilization_pct": { "min": 17.0, "median": 87.0, "max": 90.0, "mean": 77.9, "n": 14 }, "temperature_c": { "min": 38.0, "median": 43.0, "max": 43.0, "mean": 42.1, "n": 14 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 14 }, "clocks_sm_mhz": { "min": 1020.0, "median": 1548.5, "max": 1642.0, "mean": 1473.6, "n": 14 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 13286.7, "n": 14 }, "throttle_sw_power_cap_fraction": 0.8571, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 14 }, "busy_samples": 14 } } }, "energy_j_per_1k_tokens": 499.77, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 14, "temp_max_c": 43.0, "temp_min_c": 38.0, "temp_rise_c": 5.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1020.0, "clock_max_mhz": 1642.0, "clock_median_mhz": 1548.5, "mem_clock_floor_mhz": 9001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": false, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 80.75, "power_pct_of_limit": 85.0, "power_pinned_at_limit": false, "sw_power_cap_active_fraction": 0.8571, "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 85.0% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it" } }, { "level": 4, "wall_s": 4.4283, "streams": [ { "ttft_ms": 701.81, "first_token_was_thinking": false, "wall_s": 4.4274, "eval_count": 256, "eval_duration_ns": 3722034000, "decode_tok_s": 68.78, "prompt_eval_count": 536, "prompt_eval_duration_ns": 214922000, "prefill_tok_s": 2493.928, "load_duration_ns": 335116055, "total_duration_ns": 4423659901, "streamed_chunks": 256, "response_chars": 1041, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.445, "decode_tok_s_wall_gross": 57.821, "num_ctx_option": 4096, "stream": 1 }, { "ttft_ms": 701.53, "first_token_was_thinking": false, "wall_s": 4.4268, "eval_count": 256, "eval_duration_ns": 3722081000, "decode_tok_s": 68.779, "prompt_eval_count": 536, "prompt_eval_duration_ns": 245924000, "prefill_tok_s": 2179.535, "load_duration_ns": 300251739, "total_duration_ns": 4423265238, "streamed_chunks": 256, "response_chars": 1043, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.452, "decode_tok_s_wall_gross": 57.83, "num_ctx_option": 4096, "stream": 2 }, { "ttft_ms": 703.22, "first_token_was_thinking": false, "wall_s": 4.4274, "eval_count": 256, "eval_duration_ns": 3722089000, "decode_tok_s": 68.779, "prompt_eval_count": 536, "prompt_eval_duration_ns": 151810000, "prefill_tok_s": 3530.729, "load_duration_ns": 375098187, "total_duration_ns": 4424830918, "streamed_chunks": 256, "response_chars": 1041, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.471, "decode_tok_s_wall_gross": 57.821, "num_ctx_option": 4096, "stream": 3 }, { "ttft_ms": 703.03, "first_token_was_thinking": false, "wall_s": 4.4274, "eval_count": 256, "eval_duration_ns": 3722061000, "decode_tok_s": 68.779, "prompt_eval_count": 536, "prompt_eval_duration_ns": 183195000, "prefill_tok_s": 2925.844, "load_duration_ns": 343024826, "total_duration_ns": 4424002907, "streamed_chunks": 256, "response_chars": 1044, "thinking_chars": 0, "done_reason": "length", "decode_tok_s_wall": 68.468, "decode_tok_s_wall_gross": 57.822, "num_ctx_option": 4096, "stream": 4 } ], "streams_ok": 4, "streams_failed": 0, "stream_errors": [], "trial_void": false, "per_stream_decode_tok_s": [ 68.78, 68.779, 68.779, 68.779 ], "per_stream_ttft_ms": [ 701.81, 701.53, 703.22, 703.03 ], "aggregate_sum_tok_s": 275.117, "aggregate_wall_tok_s": 231.239, "tokens_decoded_total": 1024, "per_stream_decode": { "min": 68.779, "median": 68.779, "max": 68.78, "mean": 68.779, "n": 4 }, "per_stream_ttft": { "min": 701.53, "median": 702.42, "max": 703.22, "mean": 702.4, "n": 4 }, "power": { "idle_baseline": { "seconds": 3.0, "samples": 6, "power_w": { "min": 17.6, "median": 20.01, "max": 93.39, "mean": 39.09, "n": 6 }, "utilization_pct": { "min": 0.0, "median": 0.0, "max": 86.0, "mean": 14.3, "n": 6 }, "memory_used_mib": { "min": 18563.0, "median": 18563.0, "max": 18563.0, "mean": 18563.0, "n": 6 }, "power_w_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 39.09 }, "memory_used_mib_per_card": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": 18563.0 }, "power_w_all_cards": 39.09, "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." } }, "mean_w": 91.77, "max_w": 94.54, "over_idle_w": 52.68, "samples": 17, "per_gpu": { "GPU-edff232c-7dbf-2bac-07fb-921a7c9eecff": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 15.98, "median": 93.22, "max": 94.54, "mean": 71.86, "n": 17 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 17 }, "memory_used_mib": { "min": 18563.0, "median": 18563.0, "max": 18563.0, "mean": 18563.0, "n": 17 }, "utilization_pct": { "min": 0.0, "median": 86.0, "max": 91.0, "mean": 61.6, "n": 17 }, "temperature_c": { "min": 38.0, "median": 43.0, "max": 44.0, "mean": 42.4, "n": 17 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 17 }, "clocks_sm_mhz": { "min": 1215.0, "median": 1567.0, "max": 1837.0, "mean": 1519.6, "n": 17 }, "clocks_mem_mhz": { "min": 9001.0, "median": 14001.0, "max": 14001.0, "mean": 12824.5, "n": 17 }, "throttle_sw_power_cap_fraction": 0.8235, "throttle_hw_slowdown_fraction": 0.0, "throttle_sw_thermal_fraction": 0.0, "throttle_hw_thermal_fraction": 0.0, "n": 17, "busy": { "index": "0", "name": "NVIDIA GeForce RTX 5090 Laptop GPU", "power_w": { "min": 70.61, "median": 93.69, "max": 94.54, "mean": 91.77, "n": 12 }, "power_limit_w": 95.0, "power_limit_field": "enforced.power.limit", "power_limit_enforced_w": 95.0, "cap": "fixed", "power_posture": "fixed", "power_limit_enforced_stats_w": { "min": 95.0, "median": 95.0, "max": 95.0, "mean": 95.0, "n": 12 }, "memory_used_mib": { "min": 18563.0, "median": 18563.0, "max": 18563.0, "mean": 18563.0, "n": 12 }, "utilization_pct": { "min": 85.0, "median": 87.0, "max": 91.0, "mean": 87.3, "n": 12 }, "temperature_c": { "min": 42.0, "median": 44.0, "max": 44.0, "mean": 43.5, "n": 12 }, "memory_temperature_c": null, "fan_pct": null, "pcie_width_current": { "min": 8.0, "median": 8.0, "max": 8.0, "mean": 8.0, "n": 12 }, "clocks_sm_mhz": { "min": 1215.0, "median": 1447.0, "max": 1642.0, "mean": 1458.5, "n": 12 }, "clocks_mem_mhz": { "min": 14001.0, "median": 14001.0, "max": 14001.0, "mean": 14001.0, "n": 12 }, "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": 12 }, "busy_samples": 12 } } }, "energy_j_per_1k_tokens": 396.86, "thermal": { "rule_version": 2, "measured_over": "samples where the GPU was busy (utilisation > 0)", "busy_samples": 12, "temp_max_c": 44.0, "temp_min_c": 42.0, "temp_rise_c": 2.0, "memory_temp_max_c": null, "memory_temp_median_c": null, "memory_temp_readable": false, "pcie_width_min": 8.0, "pcie_width_max": 8.0, "pcie_width_median": 8.0, "fan_mean_pct": null, "fan_max_pct": null, "clock_floor_mhz": 1215.0, "clock_max_mhz": 1642.0, "clock_median_mhz": 1447.0, "mem_clock_floor_mhz": 14001.0, "mem_clock_median_mhz": 14001.0, "mem_clock_max_mhz": 14001.0, "mem_clock_held": true, "memory_bus_width_bits": 256, "memory_technology": "GDDR7", "memory_bits_per_clock": null, "peak_mem_bandwidth_gbs_at_median_clock": null, "peak_mem_bandwidth_gbs_at_floor_clock": null, "peak_mem_bandwidth_gbs_at_max_clock": null, "peak_mem_bandwidth_withheld_because": "this harness's peak-bandwidth formula assumes two bits per memory clock per pin, which is exact for GDDR6 and GDDR6X and is checked against three published bandwidths. It has no receipt for the bits-per-clock of this board's memory, so the DERIVED bytes-per-second ceiling is withheld rather than computed from an unverified constant. The memory clock itself is a reading and is published unchanged.", "peak_mem_bandwidth_formula": "GB/s = clocks.mem MHz x 1e6 x x (bus bits / 8) / 1e9; a CEILING at the clock that was observed, never an achieved rate. The bits-per-clock is registered per board and is NOT defaulted.", "clock_dropped": true, "power_limit_w": 95.0, "power_mean_w": 91.77, "power_pct_of_limit": 96.6, "power_pinned_at_limit": true, "sw_power_cap_active_fraction": 1.0, "hw_slowdown_active_fraction": 0.0, "sw_thermal_active_fraction": 0.0, "hw_thermal_active_fraction": 0.0, "at_throttling_temperature": false, "stop_core_temp_c": 83.0, "stop_memory_temp_c": 100.0, "hit_core_temp_stop": false, "hit_memory_temp_stop": false, "throttle_temperature_c": 80.0, "temp_rising_within_run": false, "verdict": "the POWER CAP doing its job: the SM clock fell while draw sat at 96.6% of the 95 W cap, at 44 C -- well below any throttling temperature" } } ], "spread_gate": { "limit": 0.15, "values": [ 275.732, 277.023, 275.117 ], "spread_fraction": 0.0069, "verdict": "spread 0.7% of the median within the 15% gate" }, "voided_by_gate": false, "derived": { "aggregate_sum_tok_s": { "min": 275.117, "median": 275.732, "max": 277.023, "mean": 275.957, "n": 3 }, "aggregate_wall_tok_s": { "min": 161.574, "median": 231.239, "max": 235.458, "mean": 209.424, "n": 3 }, "per_stream_decode_tok_s": { "min": 67.784, "median": 69.255, "max": 69.324, "mean": 68.989, "n": 12 }, "per_stream_ttft_ms": { "min": 547.32, "median": 702.42, "max": 2638.21, "mean": 1322.12, "n": 12 }, "power_mean_w": { "min": 78.34, "median": 80.75, "max": 91.77, "mean": 83.62, "n": 3 }, "energy_j_per_1k_tokens": { "min": 332.71, "median": 396.86, "max": 499.77, "mean": 409.78, "n": 3 }, "temp_max_c": { "min": 43.0, "median": 43.0, "max": 44.0, "mean": 43.3, "n": 3 }, "fan_mean_pct": null, "clock_floor_mhz": { "min": 1020.0, "median": 1215.0, "max": 1230.0, "mean": 1155.0, "n": 3 } }, "thermal_verdict": "NEITHER cap: the SM clock varied with draw at 82.5% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it; NEITHER cap: the SM clock varied with draw at 85.0% of the cap and the card at 43 C. On a memory-bound decode the clock follows the work, and nothing here was limiting it; the POWER CAP doing its job: the SM clock fell while draw sat at 96.6% of the 95 W cap, at 44 C -- well below any throttling temperature" } ], "errors": [], "aggregate_note": "aggregate_sum_tok_s is the sum of the streams' own decode rates -- what the card produced while producing. aggregate_wall_tok_s is every stream's tokens over the wall time of the whole level -- what a caller experiences, queueing and ragged finish included. Both are printed; neither is the 'real' one on its own.", "ollama_version": "0.32.13", "think_mode": false, "think_note": "think:false accepted", "ps_after_load": [ { "name": "gemma4:26b", "model": "gemma4:26b", "size": 17728054230, "digest": "5571076f3d70050487b26b341705799e0ab29b808164f90d20d4cf84f699d251", "details": { "parent_model": "", "format": "gguf", "family": "gemma4", "families": [ "gemma4" ], "parameter_size": "25.8B", "quantization_level": "Q4_K_M" }, "expires_at": "2026-09-21T19:23:34.822492318Z", "size_vram": 17728054230, "context_length": 4096 } ], "size_total": 17728054230, "size_vram": 17728054230, "fit_verdict": "fits: fully resident on the card", "ps_final": [], "finished_utc": "2026-09-21T19:15:15Z", "headline": "one-95w takes 4 parallel streams at 275.7 tok/s aggregate (sum of streams) / 231.2 tok/s over wall" }