Speed test matrix

Qwen3.5-4B-Base 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 30, 2026, 14:08 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

env SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 "$HOME/env_h200-qwen-sglang/bin/sglang" serve --model-path Qwen/Qwen3.5-4B-Base --served-model-name qwen --trust-remote-code --tp 1 --reasoning-parser qwen3 --tool-call-parser qwen3_coder --context-length 262144 --mem-fraction-static 0.863403 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length

8K context39.4K207 ms TTFT

16K context44.3K364 ms TTFT

32K context42.9K748 ms TTFT

64K context37.4K1,714 ms TTFT

128K context30K4,267 ms TTFT

251K context21.2K11,858 ms TTFT

256K context21.1K12,126 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	261.077 ms TTFT	459.798 ms TTFT	950.1115 ms TTFT	1,556.6191 ms TTFT	2,827.6353 ms TTFT	3,884.1663 ms TTFT	5,995.31,279 ms TTFT	6,315.62,464 ms TTFT
16K	255.3102 ms TTFT	100.4134 ms TTFT	743.1217 ms TTFT	1,484.2334 ms TTFT	2,339.7499 ms TTFT	2,481.41,298 ms TTFT	3,661.42,545 ms TTFT	Over capacity
32K	246.6156 ms TTFT	114.1244 ms TTFT	816.1331 ms TTFT	1,342.8561 ms TTFT	1,585.31,581 ms TTFT	2,525.93,088 ms TTFT	Over capacity	Over capacity
64K	231.7269 ms TTFT	227.1551 ms TTFT	706.0751 ms TTFT	852.71,920 ms TTFT	1,387.83,911 ms TTFT	Over capacity	Over capacity	Over capacity
128K	204.7510 ms TTFT	274.1979 ms TTFT	412.01,846 ms TTFT	716.83,744 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	165.8977 ms TTFT	218.61,904 ms TTFT	356.93,901 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	39.4K	—	207 ms	—	—	1	—	—
Prefill	16K	1	Valid	44.3K	—	364 ms	—	—	1	—	—
Prefill	32K	1	Valid	42.9K	—	748 ms	—	—	1	—	—
Prefill	64K	1	Valid	37.4K	—	1,714 ms	—	—	1	—	—
Prefill	128K	1	Valid	30K	—	4,267 ms	—	—	1	—	—
Prefill	251K	1	Valid	21.2K	—	11,858 ms	—	—	1	—	—
Prefill	256K	1	Valid	21.1K	—	12,126 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	261.0	262.9	77 ms	3.8 ms	1.0	2	7,828	30.0 s
Sustained decode	8K	2	Valid	459.7	227.4	98 ms	4.5 ms	2.3	2	13,791	30.0 s
Sustained decode	8K	4	Valid	950.1	239.6	115 ms	4.2 ms	4.0	8	28,502	30.0 s
Sustained decode	8K	8	Valid	1,556.6	199.9	191 ms	5.0 ms	10.0	8	46,698	30.0 s
Sustained decode	8K	16	Valid	2,827.6	177.5	353 ms	5.6 ms	16.8	16	84,829	30.0 s
Sustained decode	8K	32	Valid	3,884.1	120.1	663 ms	8.3 ms	38.9	0	116,510	30.0 s
Sustained decode	8K	64	Valid	5,995.3	95.1	1,279 ms	10.5 ms	64.0	0	179,872	30.0 s
Sustained decode	8K	128	Valid	6,315.6	48.9	2,464 ms	20.5 ms	157.5	0	189,470	30.0 s
Sustained decode	16K	1	Valid	255.3	257.3	102 ms	3.9 ms	1.0	2	7,660	30.0 s
Sustained decode	16K	2	Valid	100.4	50.6	134 ms	19.8 ms	130.0	0	3,012	30.0 s
Sustained decode	16K	4	Valid	743.1	190.8	217 ms	5.5 ms	30.5	4	22,290	30.0 s
Sustained decode	16K	8	Valid	1,484.2	188.0	334 ms	5.3 ms	9.0	8	44,524	30.0 s
Sustained decode	16K	16	Valid	2,339.7	150.7	499 ms	6.6 ms	18.0	16	70,186	30.0 s
Sustained decode	16K	32	Valid	2,481.4	78.9	1,298 ms	12.7 ms	48.0	0	74,433	30.0 s
Sustained decode	16K	64	Valid	3,661.4	57.8	2,545 ms	17.3 ms	84.8	0	109,824	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	246.6	249.6	156 ms	4.0 ms	1.0	2	7,398	30.0 s
Sustained decode	32K	2	Valid	114.1	57.3	244 ms	17.5 ms	65.7	0	3,422	30.0 s
Sustained decode	32K	4	Valid	816.1	205.1	331 ms	4.9 ms	4.2	4	24,483	30.0 s
Sustained decode	32K	8	Valid	1,342.8	168.3	561 ms	6.0 ms	8.2	8	40,280	30.0 s
Sustained decode	32K	16	Valid	1,585.3	99.7	1,581 ms	10.0 ms	21.3	0	47,552	30.0 s
Sustained decode	32K	32	Valid	2,525.9	78.2	3,088 ms	12.8 ms	32.9	0	75,775	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	231.7	233.8	269 ms	4.3 ms	1.0	1	6,951	30.0 s
Sustained decode	64K	2	Valid	227.1	106.5	551 ms	9.4 ms	21.9	0	6,812	30.0 s
Sustained decode	64K	4	Valid	706.0	182.2	751 ms	5.5 ms	4.0	4	21,179	30.0 s
Sustained decode	64K	8	Valid	852.7	105.9	1,920 ms	9.4 ms	9.9	0	25,576	30.0 s
Sustained decode	64K	16	Valid	1,387.8	91.3	3,911 ms	11.0 ms	16.0	0	41,632	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	204.7	208.4	510 ms	4.8 ms	1.0	1	6,141	30.0 s
Sustained decode	128K	2	Valid	274.1	153.2	979 ms	6.7 ms	6.6	2	8,222	30.0 s
Sustained decode	128K	4	Valid	412.0	105.2	1,846 ms	9.5 ms	5.8	0	12,360	30.0 s
Sustained decode	128K	8	Valid	716.8	92.8	3,744 ms	10.8 ms	8.0	0	21,504	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	165.8	171.5	977 ms	5.8 ms	1.0	1	4,974	30.0 s
Sustained decode	251K	2	Valid	218.6	103.2	1,904 ms	9.7 ms	4.1	0	6,556	30.0 s
Sustained decode	251K	4	Valid	356.9	94.1	3,901 ms	10.6 ms	4.0	0	10,708	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.