Run 906a5086

Kind	speed (llama-bench)
Started	2026-09-15 18:45 UTC
Duration	159 s
Engine	`llama.cpp@91f6a6cf3` (build 10883)
Driver	595.99.02
Throttled	no
Command	`lab bench qwen3.8-27b-q4_k_m-gguf/32k-q8kv --wait`
Lab version	`0.1.0+16df55c`

Metrics

Metric	Method	Prompt	Gen	Depth	Value	± sd	Unit
`gpu_w_avg`	llama-bench	—	128	0	247	—	W
`gpu_w_avg`	llama-bench	512	—	0	174	—	W
`gpu_w_avg`	llama-bench	—	128	4k	245	—	W
`gpu_w_avg`	llama-bench	512	—	4k	227	—	W
`gpu_w_avg`	llama-bench	—	128	16k	245	—	W
`gpu_w_avg`	llama-bench	512	—	16k	240	—	W
`pp_tps`	llama-bench	512	—	0	1,037	127	t/s
`pp_tps`	llama-bench	512	—	4k	1,091	23.3	t/s
`pp_tps`	llama-bench	512	—	16k	988	14.8	t/s
`proc_rss_peak_mb`	llama-bench	—	—	16k	16,144	—	MiB
`ram_peak_mb`	llama-bench	—	—	16k	16,256	—	MiB
`tg_tps`	llama-bench	—	128	0	31.7	0.29	t/s
`tg_tps`	llama-bench	—	128	4k	30.8	0.19	t/s
`tg_tps`	llama-bench	—	128	16k	28.2	0.10	t/s
`tokens_per_joule`	llama-bench	—	128	0	0.13	—	tok/J
`tokens_per_joule`	llama-bench	512	—	0	5.97	—	tok/J
`tokens_per_joule`	llama-bench	—	128	4k	0.13	—	tok/J
`tokens_per_joule`	llama-bench	512	—	4k	4.80	—	tok/J
`tokens_per_joule`	llama-bench	—	128	16k	0.12	—	tok/J
`tokens_per_joule`	llama-bench	512	—	16k	4.11	—	tok/J
`vram_peak_mb`	llama-bench	—	—	16k	17,268	—	MiB

Raw

{
  "bench": {
    "n_prompt": 512,
    "n_gen": 128,
    "depths": [
      0,
      4096,
      16384
    ],
    "reps": 5,
    "delay_s": 3,
    "http_prompt_lengths": [
      512,
      4096
    ]
  },
  "telemetry_summary": {
    "vram_baseline_mb": 0.9375,
    "vram_peak_mb": 17269.125,
    "ram_baseline_mb": 14852.2890625,
    "ram_peak_mb": 16256.44921875,
    "proc_rss_peak_mb": 16143.81640625,
    "temp_max_c": 64,
    "power_max_w": 252.324,
    "clock_max_mhz": 1620,
    "throttled": false,
    "power_capped_frac": 0.7097026604068858,
    "phases": {
      "llama-bench": {
        "duration_s": 128.24278868595138,
        "energy_j": 26670.47,
        "power_avg_w": 207.96857486710027
      }
    }
  }
}