Speed test matrix

Qwen3.5-35B-A3B on H200 NVL 141GB

sglang · 0.5.6.post3.dev9269+g834400705 · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 2, 2026, 00:29 UTC

Hardware

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

Engine

Engine: sglang
Version: 0.5.6.post3.dev9269+g834400705
Quantization: BF16
Backend: cuda
Repository: github.com/sgl-project/sglang
Build: cuda-13.2
Commit: 834400705f2d

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-z-lab-sglang/bin/sglang" serve --model-path Qwen/Qwen3.5-35B-A3B --served-model-name qwen --trust-remote-code --tp 1 --reasoning-parser qwen3 --tool-call-parser qwen3_coder --context-length 262144 --mem-fraction-static 0.857035 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted> --linear-attn-prefill-backend flashinfer --linear-attn-decode-backend flashinfer --mamba-scheduler-strategy extra_buffer --max-running-requests 32 --cuda-graph-max-bs-decode 32 --cuda-graph-backend-prefill tc_piecewise --enable-flashinfer-allreduce-fusion --attention-backend flashinfer --speculative-algorithm DFLASH --speculative-draft-model-path z-lab/Qwen3.5-35B-A3B-DFlash --speculative-dflash-block-size 8 --speculative-draft-attention-backend fa4

Prefill

prompt tok/s by context length

8K context28.1K290 ms TTFT

16K context32.4K497 ms TTFT

32K context28.9K1,110 ms TTFT

64K context22.8K2,806 ms TTFT

128K context15.7K8,143 ms TTFT

251K context9,96225,174 ms TTFT

256K context9,83426,011 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	358.494 ms TTFT	539.598 ms TTFT	905.0114 ms TTFT	1,464.6159 ms TTFT	2,057.0355 ms TTFT	2,945.8Requests queued1,026 ms TTFT	Over capacity	Over capacity
16K	355.1115 ms TTFT	514.3122 ms TTFT	897.9172 ms TTFT	1,412.2248 ms TTFT	1,933.0678 ms TTFT	2,716.8Requests queued1,359 ms TTFT	Over capacity	Over capacity
32K	350.3167 ms TTFT	430.0260 ms TTFT	891.2315 ms TTFT	1,367.0538 ms TTFT	1,798.41,690 ms TTFT	Over capacity	Over capacity	Over capacity
64K	350.9303 ms TTFT	505.4316 ms TTFT	778.1740 ms TTFT	1,188.21,288 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
128K	305.2572 ms TTFT	347.4774 ms TTFT	691.11,221 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
251K	259.41,006 ms TTFT	372.21,966 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

Requests queued: the server could not run all requests at once, so some waited in a queue. 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	28.1K	—	290 ms	—	—	1	—	—
Prefill	16K	1	Valid	32.4K	—	497 ms	—	—	1	—	—
Prefill	32K	1	Valid	28.9K	—	1,110 ms	—	—	1	—	—
Prefill	64K	1	Valid	22.8K	—	2,806 ms	—	—	1	—	—
Prefill	128K	1	Valid	15.7K	—	8,143 ms	—	—	1	—	—
Prefill	251K	1	Valid	9,962.0	—	25,174 ms	—	—	1	—	—
Prefill	256K	1	Valid	9,834.0	—	26,011 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	358.4	363.0	94 ms	2.8 ms	0.9	2	10,749	30.0 s
Sustained decode	8K	2	Valid	539.5	301.7	98 ms	3.3 ms	2.4	4	16,185	30.0 s
Sustained decode	8K	4	Valid	905.0	228.7	114 ms	4.4 ms	4.5	4	27,146	30.0 s
Sustained decode	8K	8	Valid	1,464.6	188.0	159 ms	5.3 ms	8.0	8	43,938	30.0 s
Sustained decode	8K	16	Valid	2,057.0	133.7	355 ms	7.5 ms	18.0	15	61,700	30.0 s
Sustained decode	8K	32	Requests queued	2,945.8	102.3	1,026 ms	9.8 ms	29.0	41	88,316	30.0 s
Sustained decode	8K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	355.1	360.9	115 ms	2.8 ms	1.0	2	10,651	30.0 s
Sustained decode	16K	2	Valid	514.3	272.1	122 ms	3.7 ms	3.1	3	15,427	30.0 s
Sustained decode	16K	4	Valid	897.9	221.2	172 ms	4.5 ms	4.3	4	26,929	30.0 s
Sustained decode	16K	8	Valid	1,412.2	179.9	248 ms	5.6 ms	7.9	8	42,355	30.0 s
Sustained decode	16K	16	Valid	1,933.0	123.8	678 ms	8.1 ms	18.8	7	57,993	30.0 s
Sustained decode	16K	32	Requests queued	2,716.8	98.0	1,359 ms	10.2 ms	28.3	49	81,451	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	350.3	364.0	167 ms	2.7 ms	1.0	2	10,508	30.0 s
Sustained decode	32K	2	Valid	430.0	228.8	260 ms	4.6 ms	6.4	2	12,899	30.0 s
Sustained decode	32K	4	Valid	891.2	223.6	315 ms	4.5 ms	3.9	4	26,727	30.0 s
Sustained decode	32K	8	Valid	1,367.0	173.4	538 ms	5.8 ms	7.9	8	41,003	30.0 s
Sustained decode	32K	16	Valid	1,798.4	113.9	1,690 ms	8.8 ms	19.7	0	53,917	30.0 s
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	350.9	360.7	303 ms	2.8 ms	1.0	2	10,532	30.0 s
Sustained decode	64K	2	Valid	505.4	265.8	316 ms	3.8 ms	2.0	4	15,157	30.0 s
Sustained decode	64K	4	Valid	778.1	200.8	740 ms	5.0 ms	4.6	4	23,320	30.0 s
Sustained decode	64K	8	Valid	1,188.2	158.4	1,288 ms	6.3 ms	8.3	8	35,640	30.0 s
Sustained decode	64K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	305.2	318.4	572 ms	3.1 ms	0.9	2	9,155	30.0 s
Sustained decode	128K	2	Valid	347.4	198.5	774 ms	5.3 ms	6.6	2	10,420	30.0 s
Sustained decode	128K	4	Valid	691.1	179.3	1,221 ms	5.6 ms	4.3	4	20,726	30.0 s
Sustained decode	128K	8	Over capacity	—	—	—	—	—	—	—	—
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	259.4	280.2	1,006 ms	3.6 ms	0.9	2	7,780	30.0 s
Sustained decode	251K	2	Valid	372.2	191.3	1,966 ms	5.2 ms	2.6	2	11,165	30.0 s
Sustained decode	251K	4	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.