{ "admission": { "ceiling_mib": 119325.0, "kv_gib": 0.3125, "kv_how": "2 x 80 layers x 8 kv heads x 64 dims x 2 bytes x 2048 tokens (an upper bound: sliding/linear layers counted full)", "line_gib": 25.0, "predicted_growth_gib": 95.82611147500575, "rule": "store bytes + KV(num_ctx, f16) + 1.0 GiB <= 25.0 GiB beside (compute M-8); the guard's own ceiling on the bench's measured memory is 26 GiB", "store_gib": 94.51361147500575 }, "argv": [ "/workshop/bench-coi-2026-09-23/coilargecard.py", "--arm-id", "f-falcon-180b-cd", "--model", "falcon:180b", "--num-ctx", "2048", "--think", "off", "--tasks", "c,d", "--mode", "full", "--ceiling-mib", "119325", "--out", "/workshop/bench-coi-2026-09-23/rows/f-falcon-180b-cd", "--expect-digest", "e2bc879d7cee", "--probes" ], "arm_id": "f-falcon-180b-cd", "baseline": { "bench_memory": { "cgroup_pids": [ 1950261 ], "per_board_mib": { "": 0.0, "": 0.0 }, "procs": [], "stamp": "2026-09-24T06:38:34Z" }, "boards": [ { "memory_mib": 2.0, "sm_mhz": 180.0, "temp_c": 35.0, "uuid": "", "watts": 19.07 }, { "memory_mib": 1.0, "sm_mhz": 210.0, "temp_c": 42.0, "uuid": "", "watts": 27.7 } ] }, "bench_unit_env": "BindsTo=coi-guard-desktop.service\nRuntimeMaxUSec=40min\nExecStart={ path=/workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/bin/ollama ; argv[]=/workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/bin/ollama serve ; ignore_errors=no ; start_time=[Thu 2026-09-24 06:38:09 UTC] ; stop_time=[n/a] ; pid=1950261 ; code=(null) ; status=0/0 }\nEnvironment=TZ=UTC OLLAMA_HOST=127.0.0.1:withheld-2026-10-14 OLLAMA_KEEP_ALIVE=-1 OLLAMA_MAX_LOADED_MODELS=1 OLLAMA_=1 OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=f16 OLLAMA_NOPRUNE=1 OLLAMA_MODELS=/workshop/bench-coi-2026-09-23/store OLLAMA_GPU_OVERHEAD=536870912 OLLAMA_LOAD_TIMEOUT=10m CUDA_VISIBLE_DEVICES=, OLLAMA_VULKAN=0 GGML_VK_VISIBLE_DEVICES=-1 HIP_VISIBLE_DEVICES= ROCR_VISIBLE_DEVICES=", "board": "", "box": "the desktop", "box_unreadable_polls": 0, "card_class": "a 96 GB workstation card (P-6: energy only, no dollar field)", "disk_floor_gib": 400.0, "disk_free_gib": 425.64856338500977, "ended": "2026-09-24T06:45:56Z", "guard_state": "active", "host": "http://127.0.0.1:withheld-2026-10-14", "idle_preload": { "board": { "cards": [ "", "" ], "joules": 1383.7959763097763, "mean_watts": 98.70156763140936, "peak_watts": 55.65, "per_card": { "": { "joules": 780.2129989385604, "mean_watts": 55.65, "min_watts": 55.65, "peak_watts": 55.65, "samples": 15, "seconds": 14.019999980926514 }, "": { "joules": 603.5829773712159, "mean_watts": 43.07, "min_watts": 43.07, "peak_watts": 43.07, "samples": 15, "seconds": 14.013999938964844 } }, "samples": 30, "seconds": 14.019999980926514 }, "box": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 3765.4979639053345, "mean_watts": 289.60913750858634, "median_s_between_changes": null, "min_watts": 161.0, "peak_watts": 337.0, "samples": 14, "seconds": 13.001999855041504, "value_changes": 1 }, "what": "the desktop's larger card + the desktop's 3090, every seat on the desktop stopped and NO bench model", "window_epoch": [ 1790231899.056, 1790231913.076 ] }, "instrument": { "coilargecard_py": "ac9f40ebcfb62b21a9d3e916f358a92119e6ee0781e8630999995f93ae41773d", "guard_py": "3c7d72c55e927c698a41fb2dfad84a9cd5d781d4f597c90d76b3a12a3fdbc16e", "jev/kit/doorman_planted.json": "4daf059c26a47a84fa3b73630a64ba1a348da72c2e5df7f3c28c007b67da22f7", "jev/kit/task_c.json": "61a0b1d6e7cf66fd91992c0c3a4572f6fbae90e6471048fd7dc1eb41ff2d7f39", "jev/run.py": "withheld-2026-10-14", "jev/run_addenda.py": "a34aca61833ed0c5860c91c7e2e227429e8ea9e2e50fec68d7b7fec15d89acab" }, "limit": null, "mode": "full", "model": "falcon:180b", "num_ctx": 2048, "ollama_version": "0.32.13", "pin": { "api_ps_model": { "context_length": 2048, "details": { "families": null, "family": "falcon", "format": "gguf", "parameter_size": "180B", "parent_model": "", "quantization_level": "Q4_0" }, "digest": "e2bc879d7cee51691d97c8425067accd534cb70df12bdc13b1bb402810702765", "expires_at": "2319-01-04T06:27:07.974773493Z", "model": "falcon:180b", "name": "falcon:180b", "size": 101958937477, "size_vram": 101958937477 }, "bench_growth_mib": 98168.0, "bench_growth_per_board_mib": { "": 78640.0, "": 19528.0 }, "bench_memory_after": { "cgroup_pids": [ 1950261, 1951645 ], "per_board_mib": { "": 78640.0, "": 19528.0 }, "procs": [ { "mib": 78640.0, "name": "/workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server", "pid": 1951645, "uuid": "" }, { "mib": 19528.0, "name": "/workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server", "pid": 1951645, "uuid": "" } ], "stamp": "2026-09-24T06:39:54Z" }, "bench_on_othercard_mib": 19528.0, "board_total_before_mib": { "": 2.0, "": 1.0 }, "board_total_growth_mib": { "": 78650.0, "": 19538.0 }, "ok": true, "ps_size": 101958937477, "ps_size_vram": 101958937477, "ps_whole": true, "rule": "PER-PROCESS: WHOLE when the bench cgroup's own memory on the desktop's larger card grew by >= 90 % of the tag's store bytes AND /api/ps reports no CPU share; none of it on the desktop's 3090; the board totals are corroboration only (four live seats share the board)", "split": true, "store_bytes": 101483217578, "whole": true }, "postflight": { "after_unload": { "api_ps": { "models": [] }, "bench_memory": { "cgroup_pids": [ 1950261 ], "per_board_mib": { "": 0.0, "": 0.0 }, "procs": [], "stamp": "2026-09-24T06:45:56Z" }, "rows": [ { "memory_mib": 2.0, "sm_mhz": 450.0, "temp_c": 63.0, "uuid": "", "watts": 70.16 }, { "memory_mib": 1.0, "sm_mhz": 360.0, "temp_c": 65.0, "uuid": "", "watts": 121.04 } ], "stamp": "2026-09-24T06:45:56Z" }, "posture_after": { "boards_on_bus": [ "0, , NVIDIA RTX PRO 6000 Blackwell Workstation Edition, withheld-2026-10-14", "1, , NVIDIA GeForce RTX 3090, withheld-2026-10-14" ], "card": { "clocks.max.sm": "3090", "clocks.mem": "13365", "clocks.sm": "2775", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "420.00", "memory.free": "18559", "memory.total": "97887", "memory.used": "78692", "name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "600.00", "power.draw": "362.02", "power.limit": "420.00", "power.max_limit": "600.00", "pstate": "P1", "temperature.gpu": "73", "utilization.gpu": "76", "uuid": "", "vbios_version": "98.02.81.00.07" }, "compute_apps": [ ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 78682 MiB", ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 19556 MiB" ], "how": "nvidia-smi -i ; enforced.power.limit READ, never set (the seats' posture)", "stamp": "2026-09-24T06:45:50Z" } }, "posture_before": { "boards_on_bus": [ "0, , NVIDIA RTX PRO 6000 Blackwell Workstation Edition, withheld-2026-10-14", "1, , NVIDIA GeForce RTX 3090, withheld-2026-10-14" ], "card": { "clocks.max.sm": "3090", "clocks.mem": "14001", "clocks.sm": "2610", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "420.00", "memory.free": "97249", "memory.total": "97887", "memory.used": "2", "name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "600.00", "power.draw": "92.19", "power.limit": "420.00", "power.max_limit": "600.00", "pstate": "P0", "temperature.gpu": "37", "utilization.gpu": "0", "uuid": "", "vbios_version": "98.02.81.00.07" }, "compute_apps": [], "how": "nvidia-smi -i ; enforced.power.limit READ, never set (the seats' posture)", "stamp": "2026-09-24T06:38:16Z" }, "posture_othercard_before": { "clocks.max.sm": "2130", "clocks.mem": "9751", "clocks.sm": "1725", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "300.00", "memory.free": "24124", "memory.total": "24576", "memory.used": "1", "name": "NVIDIA GeForce RTX 3090", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "350.00", "power.draw": "114.40", "power.limit": "300.00", "power.max_limit": "366.00", "pstate": "P0", "temperature.gpu": "46", "utilization.gpu": "0", "uuid": "", "vbios_version": "94.02.42.C0.10" }, "posture_verdict": "420.00 W enforced as run; the desktop's larger card + the desktop's 3090 on the bus", "probes": { "decode": { "board": { "cards": [ "", "" ], "joules": 37066.48112443566, "mean_watts": 617.5070987622371, "peak_watts": 424.79, "per_card": { "": { "joules": 24608.416997580527, "mean_watts": 409.9626326624667, "min_watts": 116.49, "peak_watts": 424.79, "samples": 61, "seconds": 60.026000022888184 }, "": { "joules": 12458.064126855135, "mean_watts": 207.5617558164306, "min_watts": 139.52, "peak_watts": 213.82, "samples": 61, "seconds": 60.020999908447266 } }, "samples": 122, "seconds": 60.026000022888184 }, "box": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 46055.265918016434, "mean_watts": 780.491899376434, "median_s_between_changes": 14.001999855041504, "min_watts": 773.0, "peak_watts": 831.0, "samples": 60, "seconds": 59.007999897003174, "value_changes": 5 }, "num_predict": 256, "prompt": "Describe, in about three hundred words, a lighthouse on a rocky coast during a winter storm.", "runs": [ { "decode_tok_s": 13.213113953662539, "done_reason": "length", "end_epoch": 1790232277.8440018, "eval_count": 256, "eval_s": 19.374691, "load_s": 0.28041266, "prefill_tok_s": 243.77666307252355, "prompt_eval_count": 26, "prompt_eval_s": 0.106655, "start_epoch": 1790232257.9739742, "text_head": "\nThe lighthouse stood tall and proud, a beacon of hope amidst the raging storm. The wind howled like a pack of wolves, l", "thinking_chars": 0, "total_s": 19.841002149 }, { "decode_tok_s": 13.221133424011716, "done_reason": "length", "end_epoch": 1790232297.6360636, "eval_count": 256, "eval_s": 19.362939, "load_s": 0.281910925, "prefill_tok_s": 334.6074154151063, "prompt_eval_count": 26, "prompt_eval_s": 0.077703, "start_epoch": 1790232277.844062, "text_head": "\nThe lighthouse stood tall and proud, a beacon of hope amidst the raging storm. The wind howled like a pack of wolves, l", "thinking_chars": 0, "total_s": 19.761090619 }, { "decode_tok_s": 13.221063778153443, "done_reason": "length", "end_epoch": 1790232317.3699894, "eval_count": 256, "eval_s": 19.363041, "load_s": 0.217042067, "prefill_tok_s": 331.0667990933863, "prompt_eval_count": 26, "prompt_eval_s": 0.078534, "start_epoch": 1790232297.6361184, "text_head": "\nThe lighthouse stood tall and proud, a beacon of hope amidst the raging storm. The wind howled like a pack of wolves, l", "thinking_chars": 0, "total_s": 19.704701898 } ] }, "prefill": { "board": { "cards": [ "", "" ], "joules": 6983.60397326231, "mean_watts": 536.5399474389245, "peak_watts": 442.55, "per_card": { "": { "joules": 4358.100679858923, "mean_watts": 334.82641894608184, "min_watts": 183.07, "peak_watts": 442.55, "samples": 14, "seconds": 13.016000032424927 }, "": { "joules": 2625.503293403387, "mean_watts": 201.82206785304024, "min_watts": 129.27, "peak_watts": 249.96, "samples": 14, "seconds": 13.009000062942505 } }, "samples": 28, "seconds": 13.016000032424927 }, "box": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 9010.574073553085, "mean_watts": 750.7560409421233, "median_s_between_changes": null, "min_watts": 613.0, "peak_watts": 787.0, "samples": 13, "seconds": 12.002000093460083, "value_changes": 1 }, "runs": [ { "decode_tok_s": 1000000.0, "done_reason": "length", "end_epoch": 1790232321.4572062, "eval_count": 1, "eval_s": 1e-06, "load_s": 0.267299711, "prefill_tok_s": 627.8183480271923, "prompt_eval_count": 1309, "prompt_eval_s": 2.084998, "start_epoch": 1790232319.0616767, "text_head": "\n", "thinking_chars": 0, "total_s": 2.392586314 }, { "decode_tok_s": 1000000.0, "done_reason": "length", "end_epoch": 1790232324.01831, "eval_count": 1, "eval_s": 1e-06, "load_s": 0.266636084, "prefill_tok_s": 628.5237242459158, "prompt_eval_count": 1307, "prompt_eval_s": 2.079476, "start_epoch": 1790232321.4572496, "text_head": "F", "thinking_chars": 0, "total_s": 2.559820824 }, { "decode_tok_s": 1000000.0, "done_reason": "length", "end_epoch": 1790232326.5888374, "eval_count": 1, "eval_s": 1e-06, "load_s": 0.272752241, "prefill_tok_s": 627.610140297517, "prompt_eval_count": 1307, "prompt_eval_s": 2.082503, "start_epoch": 1790232324.0183694, "text_head": "\n", "thinking_chars": 0, "total_s": 2.568963229 }, { "decode_tok_s": 1000000.0, "done_reason": "length", "end_epoch": 1790232329.1418564, "eval_count": 1, "eval_s": 1e-06, "load_s": 0.25755259, "prefill_tok_s": 628.1114222566741, "prompt_eval_count": 1307, "prompt_eval_s": 2.080841, "start_epoch": 1790232326.588894, "text_head": "F", "thinking_chars": 0, "total_s": 2.551497717 }, { "decode_tok_s": 1000000.0, "done_reason": "length", "end_epoch": 1790232331.6998975, "eval_count": 1, "eval_s": 1e-06, "load_s": 0.273547718, "prefill_tok_s": 630.7193573504584, "prompt_eval_count": 1309, "prompt_eval_s": 2.075408, "start_epoch": 1790232329.1419034, "text_head": "\n", "thinking_chars": 0, "total_s": 2.556645027 } ], "target_tokens_approx": 1747.7777777777778 }, "ttft": { "how": "the first 12 task_c prompts, run.py's body with stream true, in task order; TTFT = client clock on localhost from the request to the first chunk carrying content or thinking", "median_s": 0.9655524835688993, "p95_s": 1.0303100469755009, "runs": [ { "end_epoch": 1790232334.5443134, "eval_count": 8, "prompt_eval_count": 234, "server_load_ms": 277.441612, "server_prompt_eval_ms": 401.6, "start_epoch": 1790232333.135863, "total_s": 1.4082976880017668, "ttft_s": 0.87199528189376 }, { "end_epoch": 1790232336.020853, "eval_count": 8, "prompt_eval_count": 334, "server_load_ms": 285.916179, "server_prompt_eval_ms": 612.071, "start_epoch": 1790232334.5444238, "total_s": 1.4762819040333852, "ttft_s": 0.9381640850333497 }, { "end_epoch": 1790232337.2711067, "eval_count": 8, "prompt_eval_count": 251, "server_load_ms": 239.160681, "server_prompt_eval_ms": 403.946, "start_epoch": 1790232336.0209312, "total_s": 1.250059092999436, "ttft_s": 0.7070870829047635 }, { "end_epoch": 1790232338.8384414, "eval_count": 8, "prompt_eval_count": 356, "server_load_ms": 289.865733, "server_prompt_eval_ms": 686.347, "start_epoch": 1790232337.2712572, "total_s": 1.567013921099715, "ttft_s": 1.029149471083656 }, { "end_epoch": 1790232340.2333598, "eval_count": 6, "prompt_eval_count": 363, "server_load_ms": 264.819802, "server_prompt_eval_ms": 688.846, "start_epoch": 1790232338.838523, "total_s": 1.3947179609676823, "ttft_s": 1.008851820952259 }, { "end_epoch": 1790232341.8019156, "eval_count": 8, "prompt_eval_count": 367, "server_load_ms": 282.860201, "server_prompt_eval_ms": 688.137, "start_epoch": 1790232340.2334507, "total_s": 1.5683292270405218, "ttft_s": 1.0303100469755009 }, { "end_epoch": 1790232343.1486368, "eval_count": 8, "prompt_eval_count": 274, "server_load_ms": 286.112139, "server_prompt_eval_ms": 460.076, "start_epoch": 1790232341.8020203, "total_s": 1.346470367978327, "ttft_s": 0.8087029380258173 }, { "end_epoch": 1790232344.4272494, "eval_count": 8, "prompt_eval_count": 251, "server_load_ms": 271.799564, "server_prompt_eval_ms": 404.208, "start_epoch": 1790232343.1487348, "total_s": 1.2783800730248913, "ttft_s": 0.7356216621119529 }, { "end_epoch": 1790232345.9034283, "eval_count": 8, "prompt_eval_count": 322, "server_load_ms": 290.291502, "server_prompt_eval_ms": 606.605, "start_epoch": 1790232344.427357, "total_s": 1.4759211159544066, "ttft_s": 0.9380460259271786 }, { "end_epoch": 1790232347.4346447, "eval_count": 8, "prompt_eval_count": 362, "server_load_ms": 250.343247, "server_prompt_eval_ms": 687.01, "start_epoch": 1790232345.9035368, "total_s": 1.5309579910244793, "ttft_s": 0.992940882104449 }, { "end_epoch": 1790232349.0038815, "eval_count": 8, "prompt_eval_count": 360, "server_load_ms": 282.606302, "server_prompt_eval_ms": 687.919, "start_epoch": 1790232347.434753, "total_s": 1.5689726440468803, "ttft_s": 1.0314190230565146 }, { "end_epoch": 1790232350.5624192, "eval_count": 8, "prompt_eval_count": 356, "server_load_ms": 272.572251, "server_prompt_eval_ms": 686.845, "start_epoch": 1790232349.003983, "total_s": 1.558302205055952, "ttft_s": 1.0197266479954123 } ] } }, "show": { "capabilities": [ "completion" ], "details": { "families": null, "family": "falcon", "format": "gguf", "parameter_size": "180B", "parent_model": "", "quantization_level": "Q4_0" }, "model_info": { "falcon.attention.head_count": 232, "falcon.attention.head_count_kv": 8, "falcon.attention.layer_norm_epsilon": 1e-05, "falcon.block_count": 80, "falcon.context_length": 2048, "falcon.embedding_length": 14848, "falcon.feed_forward_length": 59392, "falcon.tensor_data_layout": "jploski", "general.architecture": "falcon", "general.file_type": 2, "general.parameter_count": 179522565120, "general.quantization_version": 2, "tokenizer.ggml.eos_token_id": 11, "tokenizer.ggml.merges": null, "tokenizer.ggml.model": "gpt2", "tokenizer.ggml.scores": null, "tokenizer.ggml.token_type": null, "tokenizer.ggml.tokens": null }, "parameters": "stop \"User:\"\nstop \"Assistant:\"", "system_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855", "template_head": "{{ .System }}\nUser: {{ .Prompt }}\nAssistant:\n", "template_sha256": "4dec76bb1a4744ce1355e7d72a5362e1113eab96a55be46753a2320b7b12c60f" }, "started": "2026-09-24T06:38:16Z", "tag": { "details": { "context_length": 2048, "embedding_length": 14848, "families": null, "family": "falcon", "format": "gguf", "parameter_size": "180B", "parent_model": "", "quantization_level": "Q4_0" }, "digest": "e2bc879d7cee51691d97c8425067accd534cb70df12bdc13b1bb402810702765", "modified_at": "2026-09-24T06:27:45.107747479Z", "name": "falcon:180b", "size": 101483217578 }, "tasks": [ "c", "d" ], "tasks_run": { "c": { "board_idle_window_epoch": [ 1790231994.447, 1790232003.458 ], "board_task_window_epoch": [ 1790232004.515, 1790232150.565 ], "box_idle_resident": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 2552.6380298137665, "mean_watts": 319.0, "median_s_between_changes": null, "min_watts": 319.0, "peak_watts": 319.0, "samples": 9, "seconds": 8.002000093460083, "value_changes": 0 }, "box_task": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 97065.19139409065, "mean_watts": 664.7435711590733, "median_s_between_changes": 6.000999927520752, "min_watts": 319.0, "peak_watts": 853.0, "samples": 147, "seconds": 146.01899981498718, "value_changes": 20 }, "energy": { "cards": [ "", "" ], "decisions": 108, "joules": 75635.51035417318, "joules_per_decision": 700.3287995756776, "mean_watts": 517.874086811827, "net_joules_per_decision": 432.5853817889943, "peak_watts": 393.35, "per_card": { "": { "joules": 46617.447170398234, "mean_watts": 319.1882724109436, "min_watts": 86.26, "peak_watts": 393.35, "samples": 147, "seconds": 146.04999995231628 }, "": { "joules": 29018.06318377495, "mean_watts": 198.6898953764647, "min_watts": 111.24, "peak_watts": 227.13, "samples": 147, "seconds": 146.04700016975403 } }, "samples": 294, "seconds": 146.04999995231628 }, "idle": { "cards": [ "", "" ], "joules": 1784.0785962510108, "mean_watts": 197.9889704238456, "peak_watts": 111.28, "per_card": { "": { "joules": 782.3042376828193, "mean_watts": 86.81658470793116, "min_watts": 86.1, "peak_watts": 89.4, "samples": 10, "seconds": 9.010999917984009 }, "": { "joules": 1001.7743585681915, "mean_watts": 111.20941052646623, "min_watts": 111.14, "peak_watts": 111.28, "samples": 10, "seconds": 9.007999897003174 } }, "samples": 20, "seconds": 9.010999917984009, "what": "the draw of these cards immediately before the arm, with the model resident and no work in flight" }, "posture_end": { "boards_on_bus": [ "0, , NVIDIA RTX PRO 6000 Blackwell Workstation Edition, withheld-2026-10-14", "1, , NVIDIA GeForce RTX 3090, withheld-2026-10-14" ], "card": { "clocks.max.sm": "3090", "clocks.mem": "13365", "clocks.sm": "2827", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "420.00", "memory.free": "18579", "memory.total": "97887", "memory.used": "78672", "name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "600.00", "power.draw": "171.84", "power.limit": "420.00", "power.max_limit": "600.00", "pstate": "P1", "temperature.gpu": "60", "utilization.gpu": "0", "uuid": "", "vbios_version": "98.02.81.00.07" }, "compute_apps": [ ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 78662 MiB", ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 19548 MiB" ], "how": "nvidia-smi -i ; enforced.power.limit READ, never set (the seats' posture)", "stamp": "2026-09-24T06:42:32Z" }, "posture_start": { "boards_on_bus": [ "0, , NVIDIA RTX PRO 6000 Blackwell Workstation Edition, withheld-2026-10-14", "1, , NVIDIA GeForce RTX 3090, withheld-2026-10-14" ], "card": { "clocks.max.sm": "3090", "clocks.mem": "13365", "clocks.sm": "2610", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "420.00", "memory.free": "18601", "memory.total": "97887", "memory.used": "78650", "name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "600.00", "power.draw": "89.40", "power.limit": "420.00", "power.max_limit": "600.00", "pstate": "P1", "temperature.gpu": "37", "utilization.gpu": "0", "uuid": "", "vbios_version": "98.02.81.00.07" }, "compute_apps": [ ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 78640 MiB", ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 19528 MiB" ], "how": "nvidia-smi -i ; enforced.power.limit READ, never set (the seats' posture)", "stamp": "2026-09-24T06:39:54Z" }, "summary": { "accuracy": 0.16666666666666666, "brier": 1.6666666666666667, "correct": 18, "errors": 0, "eval_tokens_total": 848, "floored_rows": 0, "letters_floored_total": 0, "median_seconds": 1.3492427109740674, "n": 108, "n_labelled": 108, "n_reached": 108, "p95_seconds": 1.5500756480032578, "prompt_tokens_total": 32206, "refused": 0, "refused_by_reason": {}, "reliability": [ { "accuracy": null, "correct": 0, "gap": null, "hi": 0.1, "lo": 0.0, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.2, "lo": 0.1, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.3, "lo": 0.2, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.4, "lo": 0.3, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.5, "lo": 0.4, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.6, "lo": 0.5, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.7, "lo": 0.6, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.8, "lo": 0.7, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.9, "lo": 0.8, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": 0.16666666666666666, "correct": 18, "gap": 0.8333333333333334, "hi": 1.0, "lo": 0.9, "mean_p": 1.0, "n": 108, "sum_p": 108.0 } ], "unparsed": 0 }, "wall_seconds": 146.64368445798755 }, "d": { "board_idle_window_epoch": [ 1790232152.814, 1790232161.825 ], "board_task_window_epoch": [ 1790232162.881, 1790232255.916 ], "box_idle_resident": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 5264.1709151268005, "mean_watts": 657.8569120830729, "median_s_between_changes": null, "min_watts": 296.0, "peak_watts": 682.0, "samples": 9, "seconds": 8.001999855041504, "value_changes": 1 }, "box_task": { "instrument": "box delta (CP2000 ups.realpower, integer W; includes the desktop's 3090 and the host; never wall)", "joules": 55970.83855569363, "mean_watts": 608.2993360066417, "median_s_between_changes": 14.002000093460083, "min_watts": 296.0, "peak_watts": 831.0, "samples": 93, "seconds": 92.01200008392334, "value_changes": 6 }, "energy": { "cards": [ "", "" ], "decisions": 48, "joules": 49122.30428332925, "joules_per_decision": 1023.3813392360261, "mean_watts": 527.9981121502162, "net_joules_per_decision": 573.9899550034858, "peak_watts": 376.81, "per_card": { "": { "joules": 30443.432652492524, "mean_watts": 327.22558932039755, "min_watts": 97.06, "peak_watts": 376.81, "samples": 94, "seconds": 93.03499984741211 }, "": { "joules": 18678.871630836726, "mean_watts": 200.7789971803861, "min_watts": 124.03, "peak_watts": 249.22, "samples": 94, "seconds": 93.03199982643127 } }, "samples": 188, "seconds": 93.03499984741211 }, "idle": { "cards": [ "", "" ], "joules": 2089.260604416132, "mean_watts": 231.8566827381142, "peak_watts": 171.84, "per_card": { "": { "joules": 936.9295956861972, "mean_watts": 103.97620457485847, "min_watts": 97.31, "peak_watts": 171.84, "samples": 10, "seconds": 9.011000156402588 }, "": { "joules": 1152.3310087299346, "mean_watts": 127.92306742965904, "min_watts": 124.21, "peak_watts": 157.4, "samples": 10, "seconds": 9.008000135421753 } }, "samples": 20, "seconds": 9.011000156402588, "what": "the draw of these cards immediately before the arm, with the model resident and no work in flight" }, "posture_end": { "boards_on_bus": [ "0, , NVIDIA RTX PRO 6000 Blackwell Workstation Edition, withheld-2026-10-14", "1, , NVIDIA GeForce RTX 3090, withheld-2026-10-14" ], "card": { "clocks.max.sm": "3090", "clocks.mem": "13365", "clocks.sm": "2812", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "420.00", "memory.free": "18561", "memory.total": "97887", "memory.used": "78690", "name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "600.00", "power.draw": "143.76", "power.limit": "420.00", "power.max_limit": "600.00", "pstate": "P1", "temperature.gpu": "63", "utilization.gpu": "0", "uuid": "", "vbios_version": "98.02.81.00.07" }, "compute_apps": [ ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 78680 MiB", ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 19556 MiB" ], "how": "nvidia-smi -i ; enforced.power.limit READ, never set (the seats' posture)", "stamp": "2026-09-24T06:44:17Z" }, "posture_start": { "boards_on_bus": [ "0, , NVIDIA RTX PRO 6000 Blackwell Workstation Edition, withheld-2026-10-14", "1, , NVIDIA GeForce RTX 3090, withheld-2026-10-14" ], "card": { "clocks.max.sm": "3090", "clocks.mem": "13365", "clocks.sm": "2827", "clocks_throttle_reasons.active": "0x0000000000000000", "driver_version": "595.84", "enforced.power.limit": "420.00", "memory.free": "18579", "memory.total": "97887", "memory.used": "78672", "name": "NVIDIA RTX PRO 6000 Blackwell Workstation Edition", "pci.bus_id": "withheld-2026-10-14", "pcie.link.gen.current": "4", "pcie.link.gen.max": "4", "pcie.link.width.current": "16", "pcie.link.width.max": "16", "persistence_mode": "Disabled", "power.default_limit": "600.00", "power.draw": "171.84", "power.limit": "420.00", "power.max_limit": "600.00", "pstate": "P1", "temperature.gpu": "60", "utilization.gpu": "0", "uuid": "", "vbios_version": "98.02.81.00.07" }, "compute_apps": [ ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 78662 MiB", ", 1951645, /workshop/bench-coi-2026-09-23/runtime/ollama-0.32.13/lib/ollama/llama-server, 19548 MiB" ], "how": "nvidia-smi -i ; enforced.power.limit READ, never set (the seats' posture)", "stamp": "2026-09-24T06:42:32Z" }, "summary": { "accuracy": 0.25, "brier": 0.5, "correct": 12, "errors": 0, "eval_tokens_total": 384, "floored_rows": 0, "letters_floored_total": 0, "median_seconds": 1.9401871465379372, "n": 48, "n_labelled": 48, "n_reached": 48, "p95_seconds": 1.9750142670236528, "prompt_tokens_total": 25134, "refused": 0, "refused_by_reason": {}, "reliability": [ { "accuracy": null, "correct": 0, "gap": null, "hi": 0.1, "lo": 0.0, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.2, "lo": 0.1, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.3, "lo": 0.2, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.4, "lo": 0.3, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.5, "lo": 0.4, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": 0.25, "correct": 12, "gap": 0.25, "hi": 0.6, "lo": 0.5, "mean_p": 0.5, "n": 48, "sum_p": 24.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.7, "lo": 0.6, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.8, "lo": 0.7, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 0.9, "lo": 0.8, "mean_p": null, "n": 0, "sum_p": 0.0 }, { "accuracy": null, "correct": 0, "gap": null, "hi": 1.0, "lo": 0.9, "mean_p": null, "n": 0, "sum_p": 0.0 } ], "unparsed": 48 }, "wall_seconds": 93.1491193890106 } }, "think": "off", "think_budget": 8, "ups_before": { "available": true, "load_pct": 28, "read_utc": "2026-09-24T06:38:16Z", "ups.realpower": 337, "ups.status": "OL" }, "warm": { "done_reason": "length", "load_ms": 76629.946547, "seconds": 76.8327767449664 } }