Speed test matrix

Qwen3.6-27B-int4-AutoRound on H200 NVL 141GB

vllm · 0.26.1rc1.dev638+g52be12cfa · INT4+BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 4, 2026, 00:26 UTC

Hardware

Type: Discrete GPU
GPU: H200 NVL
VRAM: 141 GB
OS: linux

Engine

Engine: vllm
Version: 0.26.1rc1.dev638+g52be12cfa
Quantization: INT4+BF16
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2
Commit: 52be12cfac0c

Run

Served model: qwen
Revision: main
Benchmark tool: llm-inference-bench v0.4.29
Decode test: Fixed duration
Prefill mode: Measured during decode warm-up
Time per measurement: 30 s
Max output: 4,096 tokens

Launch command

"$HOME/env_h200-intel-vllm/bin/vllm" serve Intel/Qwen3.6-27B-int4-AutoRound --served-model-name qwen --trust-remote-code --tensor-parallel-size 1 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.883581 --host 0.0.0.0 --port 8000 --api-key <redacted> --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'

Prefill

prompt tok/s by context length

8K context9,975821 ms TTFT

16K context9,6881,674 ms TTFT

32K context9,2863,477 ms TTFT

64K context8,5197,562 ms TTFT

128K context7,23617,786 ms TTFT

251K context5,58845,136 ms TTFT

256K context5,55646,298 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests

Context	1	2	4	8	16	32	64	128
8K	144.3125 ms TTFT	277.8395 ms TTFT	534.7338 ms TTFT	972.2721 ms TTFT	1,573.21,758 ms TTFT	2,370.73,011 ms TTFT	2,879.64,804 ms TTFT	Over capacity
16K	143.1158 ms TTFT	282.5202 ms TTFT	512.6259 ms TTFT	952.2749 ms TTFT	1,462.61,683 ms TTFT	2,093.33,162 ms TTFT	2,505.24,756 ms TTFT	Over capacity
32K	146.0218 ms TTFT	263.1294 ms TTFT	502.5400 ms TTFT	870.51,199 ms TTFT	1,290.81,921 ms TTFT	1,700.82,911 ms TTFT	Over capacity	Over capacity
64K	131.9300 ms TTFT	254.0404 ms TTFT	448.6806 ms TTFT	752.91,364 ms TTFT	1,018.52,750 ms TTFT	Over capacity	Over capacity	Over capacity
128K	130.0536 ms TTFT	232.1690 ms TTFT	382.11,330 ms TTFT	576.82,379 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	115.3796 ms TTFT	193.11,082 ms TTFT	295.41,522 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

Over capacity: not run because this many requests at this context would exceed the server's KV cache.

TTFT is measured after the benchmark warms the server's prompt cache, so it mostly excludes prefill.

All measurements

Latencies are medians. Total tok/s is output across all simultaneous requests.

Phase	Context	Requests at once	Status	Total tok/s	Per request tok/s	TTFT	Time between tokens	Average active requests	Completed requests	Output tokens	Measured for
Prefill	8K	1	Valid	9,975.0	—	821 ms	—	—	1	—	—
Prefill	16K	1	Valid	9,688.0	—	1,674 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,286.0	—	3,477 ms	—	—	1	—	—
Prefill	64K	1	Valid	8,519.0	—	7,562 ms	—	—	1	—	—
Prefill	128K	1	Valid	7,236.0	—	17,786 ms	—	—	1	—	—
Prefill	251K	1	Valid	5,588.0	—	45,136 ms	—	—	1	—	—
Prefill	256K	1	Valid	5,556.0	—	46,298 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	144.3	147.5	125 ms	6.8 ms	1.0	1	4,328	30.0 s
Sustained decode	8K	2	Valid	277.8	143.3	395 ms	7.0 ms	2.0	2	8,333	30.0 s
Sustained decode	8K	4	Valid	534.7	135.0	338 ms	7.4 ms	3.9	4	16,034	30.0 s
Sustained decode	8K	8	Valid	972.2	122.1	721 ms	8.2 ms	7.9	7	29,178	30.0 s
Sustained decode	8K	16	Valid	1,573.2	99.2	1,758 ms	10.1 ms	16.0	0	47,169	30.0 s
Sustained decode	8K	32	Valid	2,370.7	73.7	3,011 ms	13.6 ms	32.0	0	71,071	30.0 s
Sustained decode	8K	64	Valid	2,879.6	43.5	4,804 ms	23.0 ms	64.0	0	86,307	30.0 s
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	143.1	148.3	158 ms	6.8 ms	1.0	1	4,293	30.0 s
Sustained decode	16K	2	Valid	282.5	144.8	202 ms	6.9 ms	2.0	2	8,474	30.0 s
Sustained decode	16K	4	Valid	512.6	130.9	259 ms	7.6 ms	4.0	4	15,386	30.0 s
Sustained decode	16K	8	Valid	952.2	120.0	749 ms	8.3 ms	8.0	2	28,541	30.0 s
Sustained decode	16K	16	Valid	1,462.6	92.0	1,683 ms	10.9 ms	16.0	0	43,845	30.0 s
Sustained decode	16K	32	Valid	2,093.3	64.9	3,162 ms	15.4 ms	32.0	0	62,749	30.0 s
Sustained decode	16K	64	Valid	2,505.2	37.7	4,756 ms	26.5 ms	64.0	0	75,096	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	146.0	148.7	218 ms	6.7 ms	1.0	1	4,380	30.0 s
Sustained decode	32K	2	Valid	263.1	139.6	294 ms	7.2 ms	2.0	2	7,890	30.0 s
Sustained decode	32K	4	Valid	502.5	126.0	400 ms	7.9 ms	4.0	4	15,028	29.9 s
Sustained decode	32K	8	Valid	870.5	109.7	1,199 ms	9.1 ms	8.0	0	26,103	30.0 s
Sustained decode	32K	16	Valid	1,290.8	80.7	1,921 ms	12.4 ms	16.0	0	38,713	30.0 s
Sustained decode	32K	32	Valid	1,700.8	52.2	2,911 ms	19.2 ms	32.0	0	50,970	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	131.9	146.5	300 ms	6.9 ms	1.0	1	3,956	30.0 s
Sustained decode	64K	2	Valid	254.0	136.2	404 ms	7.4 ms	2.0	2	7,620	30.0 s
Sustained decode	64K	4	Valid	448.6	113.0	806 ms	8.9 ms	4.0	0	13,454	30.0 s
Sustained decode	64K	8	Valid	752.9	94.2	1,364 ms	10.6 ms	8.0	0	22,585	30.0 s
Sustained decode	64K	16	Valid	1,018.5	63.3	2,750 ms	15.8 ms	16.0	0	30,555	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	130.0	140.4	536 ms	7.1 ms	1.0	1	3,901	30.0 s
Sustained decode	128K	2	Valid	232.1	116.5	690 ms	8.6 ms	2.0	0	6,962	30.0 s
Sustained decode	128K	4	Valid	382.1	95.5	1,330 ms	10.5 ms	4.0	0	11,458	30.0 s
Sustained decode	128K	8	Valid	576.8	72.2	2,379 ms	13.8 ms	8.0	0	17,315	30.0 s
Sustained decode	128K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	1	Valid	115.3	115.9	796 ms	8.6 ms	1.0	0	3,458	30.0 s
Sustained decode	251K	2	Valid	193.1	96.7	1,082 ms	10.3 ms	2.0	0	5,790	30.0 s
Sustained decode	251K	4	Valid	295.4	74.1	1,522 ms	13.5 ms	4.0	0	8,862	30.0 s
Sustained decode	251K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	128	Over capacity	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.