Speed test matrix

Qwen3.6-35B-A3B on H200 NVL 141GB

sglang · 0.5.6.post3.dev9160+g1cf2b8c54 · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Aug 31, 2026, 00:23 UTC

Hardware

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

Engine

Engine: sglang
Version: 0.5.6.post3.dev9160+g1cf2b8c54
Quantization: BF16
Backend: cuda
Repository: github.com/sgl-project/sglang
Build: cuda-13.2
Commit: 1cf2b8c54d81

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-qwen-sglang/bin/sglang" serve --model-path Qwen/Qwen3.6-35B-A3B --served-model-name qwen --trust-remote-code --tp 1 --reasoning-parser qwen3 --tool-call-parser qwen3_coder --context-length 262144 --mem-fraction-static 0.861674 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length

8K context34K240 ms TTFT

16K context37K436 ms TTFT

32K context35.6K903 ms TTFT

64K context30.9K2,072 ms TTFT

128K context24.5K5,220 ms TTFT

251K context17.2K14,563 ms TTFT

256K context17.2K14,895 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	215.479 ms TTFT	322.1103 ms TTFT	617.8133 ms TTFT	956.5201 ms TTFT	1,588.2353 ms TTFT	2,540.8667 ms TTFT	3,623.21,277 ms TTFT	Over capacity
16K	213.0105 ms TTFT	177.9137 ms TTFT	628.0216 ms TTFT	909.6354 ms TTFT	1,527.5668 ms TTFT	2,277.81,307 ms TTFT	3,211.42,501 ms TTFT	Over capacity
32K	207.6165 ms TTFT	128.5261 ms TTFT	565.3401 ms TTFT	795.3827 ms TTFT	1,388.21,556 ms TTFT	1,769.73,275 ms TTFT	Over capacity	Over capacity
64K	200.1300 ms TTFT	92.5509 ms TTFT	469.9978 ms TTFT	894.52,015 ms TTFT	1,192.63,777 ms TTFT	Over capacity	Over capacity	Over capacity
128K	186.0531 ms TTFT	172.91,043 ms TTFT	480.11,924 ms TTFT	728.53,724 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	163.51,017 ms TTFT	239.11,873 ms TTFT	392.83,697 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	34K	—	240 ms	—	—	1	—	—
Prefill	16K	1	Valid	37K	—	436 ms	—	—	1	—	—
Prefill	32K	1	Valid	35.6K	—	903 ms	—	—	1	—	—
Prefill	64K	1	Valid	30.9K	—	2,072 ms	—	—	1	—	—
Prefill	128K	1	Valid	24.5K	—	5,220 ms	—	—	1	—	—
Prefill	251K	1	Valid	17.2K	—	14,563 ms	—	—	1	—	—
Prefill	256K	1	Valid	17.2K	—	14,895 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	215.4	216.0	79 ms	4.6 ms	1.0	1	6,462	30.0 s
Sustained decode	8K	2	Valid	322.1	166.3	103 ms	6.1 ms	2.3	2	9,661	30.0 s
Sustained decode	8K	4	Valid	617.8	157.3	133 ms	6.4 ms	4.8	4	18,534	30.0 s
Sustained decode	8K	8	Valid	956.5	118.7	201 ms	8.4 ms	10.4	0	28,696	30.0 s
Sustained decode	8K	16	Valid	1,588.2	100.5	353 ms	9.9 ms	16.0	0	47,648	30.0 s
Sustained decode	8K	32	Valid	2,540.8	77.8	667 ms	12.8 ms	33.6	0	76,225	30.0 s
Sustained decode	8K	64	Valid	3,623.2	51.7	1,277 ms	19.3 ms	64.0	0	108,704	30.0 s
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	213.0	213.8	105 ms	4.7 ms	1.0	1	6,390	30.0 s
Sustained decode	16K	2	Valid	177.9	83.7	137 ms	11.9 ms	32.4	0	5,336	30.0 s
Sustained decode	16K	4	Valid	628.0	156.5	216 ms	6.4 ms	4.1	4	18,841	30.0 s
Sustained decode	16K	8	Valid	909.6	112.8	354 ms	8.9 ms	10.5	0	27,288	30.0 s
Sustained decode	16K	16	Valid	1,527.5	96.2	668 ms	10.4 ms	16.0	0	45,824	30.0 s
Sustained decode	16K	32	Valid	2,277.8	70.1	1,307 ms	14.3 ms	35.6	0	68,319	30.0 s
Sustained decode	16K	64	Valid	3,211.4	45.3	2,501 ms	22.1 ms	64.0	0	96,352	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	207.6	208.8	165 ms	4.8 ms	1.0	1	6,227	30.0 s
Sustained decode	32K	2	Valid	128.5	61.3	261 ms	16.3 ms	39.5	0	3,854	30.0 s
Sustained decode	32K	4	Valid	565.3	146.3	401 ms	6.9 ms	4.6	4	16,958	30.0 s
Sustained decode	32K	8	Valid	795.3	99.7	827 ms	10.0 ms	11.5	0	23,856	30.0 s
Sustained decode	32K	16	Valid	1,388.2	85.2	1,556 ms	11.7 ms	16.5	0	41,648	30.0 s
Sustained decode	32K	32	Valid	1,769.7	56.0	3,275 ms	17.9 ms	40.1	0	53,088	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	200.1	202.2	300 ms	4.9 ms	1.0	1	6,004	30.0 s
Sustained decode	64K	2	Valid	92.5	46.5	509 ms	21.5 ms	34.0	0	2,776	30.0 s
Sustained decode	64K	4	Valid	469.9	109.1	978 ms	9.2 ms	7.9	0	14,096	30.0 s
Sustained decode	64K	8	Valid	894.5	114.3	2,015 ms	8.8 ms	8.0	0	26,832	30.0 s
Sustained decode	64K	16	Valid	1,192.6	78.3	3,777 ms	12.8 ms	16.0	0	35,777	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	186.0	189.5	531 ms	5.3 ms	1.0	1	5,580	30.0 s
Sustained decode	128K	2	Valid	172.9	83.0	1,043 ms	12.1 ms	12.0	0	5,188	30.0 s
Sustained decode	128K	4	Valid	480.1	119.2	1,924 ms	8.4 ms	4.2	0	14,404	30.0 s
Sustained decode	128K	8	Valid	728.5	95.5	3,724 ms	10.5 ms	8.0	0	21,856	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	163.5	169.2	1,017 ms	5.9 ms	1.0	1	4,905	30.0 s
Sustained decode	251K	2	Valid	239.1	112.8	1,873 ms	8.9 ms	3.1	0	7,172	30.0 s
Sustained decode	251K	4	Valid	392.8	102.8	3,697 ms	9.7 ms	4.0	0	11,785	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.