Speed test matrix

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

sglang · 0.5.6.post3.dev9265+g05c584c44 · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 1, 2026, 17:52 UTC

Hardware

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

Engine

Engine: sglang
Version: 0.5.6.post3.dev9265+g05c584c44
Quantization: BF16
Backend: cuda
Repository: github.com/sgl-project/sglang
Build: cuda-13.2
Commit: 05c584c44fb0

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

env SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 "$HOME/env_h200-redhatai-sglang/bin/sglang" serve --model-path RedHatAI/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 context34.3K238 ms TTFT

16K context37.2K434 ms TTFT

32K context35.9K895 ms TTFT

64K context30.8K2,080 ms TTFT

128K context24.3K5,260 ms TTFT

251K context17.1K14,629 ms TTFT

256K context17.1K14,956 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	216.176 ms TTFT	323.0101 ms TTFT	619.9143 ms TTFT	958.8207 ms TTFT	1,633.4308 ms TTFT	2,570.3648 ms TTFT	3,622.21,243 ms TTFT	Over capacity
16K	213.0103 ms TTFT	172.6151 ms TTFT	624.4196 ms TTFT	923.9364 ms TTFT	1,518.9658 ms TTFT	2,275.11,332 ms TTFT	3,220.02,485 ms TTFT	Over capacity
32K	208.1167 ms TTFT	128.1271 ms TTFT	567.5392 ms TTFT	795.9824 ms TTFT	1,385.41,723 ms TTFT	1,744.53,158 ms TTFT	Over capacity	Over capacity
64K	200.9291 ms TTFT	91.6513 ms TTFT	460.3948 ms TTFT	889.62,043 ms TTFT	1,195.73,721 ms TTFT	Over capacity	Over capacity	Over capacity
128K	185.9553 ms TTFT	188.9995 ms TTFT	491.41,872 ms TTFT	569.83,720 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	163.51,011 ms TTFT	150.21,969 ms TTFT	362.53,772 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	34.3K	—	238 ms	—	—	1	—	—
Prefill	16K	1	Valid	37.2K	—	434 ms	—	—	1	—	—
Prefill	32K	1	Valid	35.9K	—	895 ms	—	—	1	—	—
Prefill	64K	1	Valid	30.8K	—	2,080 ms	—	—	1	—	—
Prefill	128K	1	Valid	24.3K	—	5,260 ms	—	—	1	—	—
Prefill	251K	1	Valid	17.1K	—	14,629 ms	—	—	1	—	—
Prefill	256K	1	Valid	17.1K	—	14,956 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	216.1	216.6	76 ms	4.6 ms	1.0	1	6,482	30.0 s
Sustained decode	8K	2	Valid	323.0	166.3	101 ms	6.1 ms	2.4	2	9,690	30.0 s
Sustained decode	8K	4	Valid	619.9	157.5	143 ms	6.4 ms	4.7	4	18,595	30.0 s
Sustained decode	8K	8	Valid	958.8	119.0	207 ms	8.4 ms	10.4	0	28,752	30.0 s
Sustained decode	8K	16	Valid	1,633.4	103.0	308 ms	9.7 ms	16.0	0	48,992	30.0 s
Sustained decode	8K	32	Valid	2,570.3	78.7	648 ms	12.7 ms	32.5	0	77,120	30.0 s
Sustained decode	8K	64	Valid	3,622.2	51.8	1,243 ms	19.3 ms	64.0	0	108,672	30.0 s
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	213.0	213.9	103 ms	4.7 ms	1.0	1	6,389	30.0 s
Sustained decode	16K	2	Valid	172.6	81.5	151 ms	12.3 ms	34.5	0	5,178	30.0 s
Sustained decode	16K	4	Valid	624.4	155.9	196 ms	6.4 ms	4.1	4	18,731	30.0 s
Sustained decode	16K	8	Valid	923.9	114.4	364 ms	8.7 ms	10.5	0	27,712	30.0 s
Sustained decode	16K	16	Valid	1,518.9	95.8	658 ms	10.4 ms	16.0	0	45,568	30.0 s
Sustained decode	16K	32	Valid	2,275.1	70.0	1,332 ms	14.3 ms	35.7	0	68,256	30.0 s
Sustained decode	16K	64	Valid	3,220.0	45.3	2,485 ms	22.1 ms	64.0	0	96,576	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	208.1	209.4	167 ms	4.8 ms	1.0	1	6,243	30.0 s
Sustained decode	32K	2	Valid	128.1	61.1	271 ms	16.4 ms	39.3	0	3,841	30.0 s
Sustained decode	32K	4	Valid	567.5	146.6	392 ms	6.8 ms	4.6	4	17,024	30.0 s
Sustained decode	32K	8	Valid	795.9	99.9	824 ms	10.0 ms	11.5	0	23,872	30.0 s
Sustained decode	32K	16	Valid	1,385.4	85.3	1,723 ms	11.7 ms	16.4	0	41,552	30.0 s
Sustained decode	32K	32	Valid	1,744.5	55.2	3,158 ms	18.1 ms	40.4	0	52,320	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	200.9	203.1	291 ms	4.9 ms	1.0	1	6,028	30.0 s
Sustained decode	64K	2	Valid	91.6	46.0	513 ms	21.7 ms	34.0	0	2,746	30.0 s
Sustained decode	64K	4	Valid	460.3	106.9	948 ms	9.4 ms	9.2	0	13,808	30.0 s
Sustained decode	64K	8	Valid	889.6	114.0	2,043 ms	8.8 ms	8.0	0	26,688	30.0 s
Sustained decode	64K	16	Valid	1,195.7	78.4	3,721 ms	12.8 ms	16.0	0	35,872	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	185.9	189.7	553 ms	5.3 ms	1.0	1	5,578	30.0 s
Sustained decode	128K	2	Valid	188.9	89.7	995 ms	11.2 ms	10.8	0	5,666	30.0 s
Sustained decode	128K	4	Valid	491.4	122.8	1,872 ms	8.1 ms	4.0	2	14,555	29.6 s
Sustained decode	128K	8	Valid	569.8	74.9	3,720 ms	13.3 ms	10.0	0	17,088	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	1,011 ms	5.9 ms	1.0	1	4,905	30.0 s
Sustained decode	251K	2	Valid	150.2	72.6	1,969 ms	13.8 ms	7.9	0	4,504	30.0 s
Sustained decode	251K	4	Valid	362.5	91.7	3,772 ms	10.9 ms	4.6	0	10,876	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.