Speed test matrix

Qwen3.5-9B-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, 16:37 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/Qwen3.5-9B-Base --served-model-name qwen --trust-remote-code --tp 1 --reasoning-parser qwen3 --tool-call-parser qwen3_coder --context-length 262144 --mem-fraction-static 0.862486 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length

8K context27.6K295 ms TTFT

16K context28.9K557 ms TTFT

32K context28.8K1,116 ms TTFT

64K context26K2,461 ms TTFT

128K context22.2K5,770 ms TTFT

251K context17K14,748 ms TTFT

256K context17.1K14,996 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	173.076 ms TTFT	264.696 ms TTFT	637.7126 ms TTFT	1,151.1210 ms TTFT	1,915.0359 ms TTFT	2,750.3675 ms TTFT	4,442.51,286 ms TTFT	4,456.92,475 ms TTFT
16K	171.0101 ms TTFT	87.7143 ms TTFT	198.1212 ms TTFT	1,049.0321 ms TTFT	1,554.7685 ms TTFT	2,920.41,274 ms TTFT	3,481.72,551 ms TTFT	Over capacity
32K	166.7159 ms TTFT	102.1275 ms TTFT	548.9432 ms TTFT	758.1821 ms TTFT	1,599.21,721 ms TTFT	2,133.83,276 ms TTFT	Over capacity	Over capacity
64K	158.8297 ms TTFT	100.3564 ms TTFT	491.5746 ms TTFT	546.31,889 ms TTFT	829.33,668 ms TTFT	Over capacity	Over capacity	Over capacity
128K	145.4534 ms TTFT	100.2980 ms TTFT	359.61,895 ms TTFT	532.53,796 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	124.6974 ms TTFT	84.91,974 ms TTFT	223.63,940 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	27.6K	—	295 ms	—	—	1	—	—
Prefill	16K	1	Valid	28.9K	—	557 ms	—	—	1	—	—
Prefill	32K	1	Valid	28.8K	—	1,116 ms	—	—	1	—	—
Prefill	64K	1	Valid	26K	—	2,461 ms	—	—	1	—	—
Prefill	128K	1	Valid	22.2K	—	5,770 ms	—	—	1	—	—
Prefill	251K	1	Valid	17K	—	14,748 ms	—	—	1	—	—
Prefill	256K	1	Valid	17.1K	—	14,996 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	173.0	173.6	76 ms	5.8 ms	1.0	1	5,190	30.0 s
Sustained decode	8K	2	Valid	264.6	148.9	96 ms	6.8 ms	2.8	2	7,937	30.0 s
Sustained decode	8K	4	Valid	637.7	161.5	126 ms	6.2 ms	5.1	4	19,130	30.0 s
Sustained decode	8K	8	Valid	1,151.1	147.7	210 ms	6.8 ms	9.6	8	34,532	30.0 s
Sustained decode	8K	16	Valid	1,915.0	121.1	359 ms	8.3 ms	21.0	16	57,441	30.0 s
Sustained decode	8K	32	Valid	2,750.3	86.7	675 ms	11.5 ms	48.0	0	82,496	30.0 s
Sustained decode	8K	64	Valid	4,442.5	67.9	1,286 ms	14.7 ms	81.6	0	133,248	30.0 s
Sustained decode	8K	128	Valid	4,456.9	35.4	2,475 ms	28.2 ms	192.0	0	133,632	30.0 s
Sustained decode	16K	1	Valid	171.0	171.8	101 ms	5.8 ms	1.0	1	5,131	30.0 s
Sustained decode	16K	2	Valid	87.7	42.4	143 ms	23.6 ms	142.8	0	2,628	30.0 s
Sustained decode	16K	4	Valid	198.1	48.4	212 ms	20.7 ms	121.2	0	5,944	30.0 s
Sustained decode	16K	8	Valid	1,049.0	136.2	321 ms	7.4 ms	9.7	8	31,470	30.0 s
Sustained decode	16K	16	Valid	1,554.7	98.1	685 ms	10.2 ms	24.0	0	46,640	30.0 s
Sustained decode	16K	32	Valid	2,920.4	89.7	1,274 ms	11.2 ms	33.1	0	87,616	30.0 s
Sustained decode	16K	64	Valid	3,481.7	54.6	2,551 ms	18.3 ms	78.4	0	104,448	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	166.7	167.7	159 ms	6.0 ms	1.0	1	5,002	30.0 s
Sustained decode	32K	2	Valid	102.1	51.2	275 ms	19.5 ms	66.0	0	3,060	30.0 s
Sustained decode	32K	4	Valid	548.9	137.9	432 ms	7.3 ms	7.8	4	16,466	30.0 s
Sustained decode	32K	8	Valid	758.1	95.9	821 ms	10.4 ms	12.0	0	22,744	30.0 s
Sustained decode	32K	16	Valid	1,599.2	98.9	1,721 ms	10.1 ms	16.6	0	47,968	30.0 s
Sustained decode	32K	32	Valid	2,133.8	66.6	3,276 ms	15.0 ms	34.3	0	64,000	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	158.8	160.6	297 ms	6.2 ms	1.0	1	4,765	30.0 s
Sustained decode	64K	2	Valid	100.3	50.2	564 ms	19.9 ms	33.3	0	3,009	30.0 s
Sustained decode	64K	4	Valid	491.5	129.0	746 ms	7.8 ms	4.8	4	14,745	30.0 s
Sustained decode	64K	8	Valid	546.3	70.3	1,889 ms	14.2 ms	12.0	0	16,384	30.0 s
Sustained decode	64K	16	Valid	829.3	53.7	3,668 ms	18.6 ms	24.6	0	24,880	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	145.4	148.3	534 ms	6.7 ms	1.0	1	4,362	30.0 s
Sustained decode	128K	2	Valid	100.2	50.4	980 ms	19.8 ms	18.0	0	3,004	30.0 s
Sustained decode	128K	4	Valid	359.6	88.3	1,895 ms	11.3 ms	5.5	0	10,788	30.0 s
Sustained decode	128K	8	Valid	532.5	66.8	3,796 ms	15.0 ms	9.3	0	15,968	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	124.6	129.4	974 ms	7.7 ms	1.0	1	3,737	30.0 s
Sustained decode	251K	2	Valid	84.9	43.1	1,974 ms	23.2 ms	10.0	0	2,546	30.0 s
Sustained decode	251K	4	Valid	223.6	58.7	3,940 ms	17.0 ms	6.0	0	6,704	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.