Run 768ae1e3

Kind	speed (llama-bench)
Started	2026-09-15 18:42 UTC
Duration	135 s
Engine	`llama.cpp@91f6a6cf3` (build 10883)
Driver	595.99.02
Throttled	no
Command	`lab bench qwen3.8-27b-q4_k_m-gguf/32k-f16kv --wait`
Lab version	`0.1.0+16df55c`

Metrics

Metric	Method	Prompt	Gen	Depth	Value	± sd	Unit
`gpu_w_avg`	llama-bench	—	128	0	245	—	W
`gpu_w_avg`	llama-bench	512	—	0	173	—	W
`gpu_w_avg`	llama-bench	—	128	4k	244	—	W
`gpu_w_avg`	llama-bench	512	—	4k	233	—	W
`gpu_w_avg`	llama-bench	—	128	16k	242	—	W
`gpu_w_avg`	llama-bench	512	—	16k	237	—	W
`pp_tps`	llama-bench	512	—	0	965	71.7	t/s
`pp_tps`	llama-bench	512	—	4k	1,109	26.2	t/s
`pp_tps`	llama-bench	512	—	16k	998	15.8	t/s
`proc_rss_peak_mb`	llama-bench	—	—	16k	16,063	—	MiB
`ram_peak_mb`	llama-bench	—	—	16k	16,256	—	MiB
`tg_tps`	llama-bench	—	128	0	32.4	0.20	t/s
`tg_tps`	llama-bench	—	128	4k	31.7	0.10	t/s
`tg_tps`	llama-bench	—	128	16k	30.3	0.16	t/s
`tokens_per_joule`	llama-bench	—	128	0	0.13	—	tok/J
`tokens_per_joule`	llama-bench	512	—	0	5.58	—	tok/J
`tokens_per_joule`	llama-bench	—	128	4k	0.13	—	tok/J
`tokens_per_joule`	llama-bench	512	—	4k	4.75	—	tok/J
`tokens_per_joule`	llama-bench	—	128	16k	0.13	—	tok/J
`tokens_per_joule`	llama-bench	512	—	16k	4.21	—	tok/J
`vram_peak_mb`	llama-bench	—	—	16k	17,762	—	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": 17763.125,
    "ram_baseline_mb": 15567.5234375,
    "ram_peak_mb": 16256.37890625,
    "proc_rss_peak_mb": 16062.8203125,
    "temp_max_c": 62,
    "power_max_w": 250.849,
    "clock_max_mhz": 1620,
    "throttled": false,
    "power_capped_frac": 0.7150928167877321,
    "phases": {
      "llama-bench": {
        "duration_s": 124.41050228994573,
        "energy_j": 25794.996,
        "power_avg_w": 207.33776912083596
      }
    }
  }
}