Speed test matrix

Laguna-XS.2-INT4 on H200 NVL 141GB

sglang · 0.5.6.post3.dev9269+g834400705 · INT4+INT86 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Aug 29, 2026, 11:19 UTC

Hardware

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

Engine

Engine: sglang
Version: 0.5.6.post3.dev9269+g834400705
Quantization: INT4+INT8
Backend: cuda
Repository: github.com/sgl-project/sglang
Build: cuda-13.2
Commit: 834400705f2d

Run

Served model: poolside
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-poolside-sglang/bin/sglang" serve --model-path poolside/Laguna-XS.2-INT4 --served-model-name poolside --trust-remote-code --tp 1 --reasoning-parser poolside_v1 --tool-call-parser poolside_v1 --context-length 262144 --mem-fraction-static 0.851767 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted> --page-size 1 --attention-backend fa3

Prefill

prompt tok/s by context length

8K context32.9K250 ms TTFT

16K context32.7K495 ms TTFT

32K context31.1K1,036 ms TTFT

64K context25.6K2,510 ms TTFT

128K context18.5K6,931 ms TTFT

251K context12.1K20,748 ms TTFT

256K context12K21,323 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	185.766 ms TTFT	228.987 ms TTFT	611.497 ms TTFT	962.7132 ms TTFT	1,447.3201 ms TTFT	2,619.0341 ms TTFT	Over capacity	Over capacity
16K	182.784 ms TTFT	249.0105 ms TTFT	590.7144 ms TTFT	886.8186 ms TTFT	1,606.8326 ms TTFT	2,404.3613 ms TTFT	Over capacity	Over capacity
32K	177.0108 ms TTFT	190.4136 ms TTFT	553.1202 ms TTFT	719.7368 ms TTFT	1,313.1687 ms TTFT	Over capacity	Over capacity	Over capacity
64K	167.3161 ms TTFT	211.9237 ms TTFT	502.9363 ms TTFT	564.8751 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
128K	151.0258 ms TTFT	179.6417 ms TTFT	375.7777 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
251K	126.6476 ms TTFT	195.9885 ms TTFT	Over capacity	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	32.9K	—	250 ms	—	—	1	—	—
Prefill	16K	1	Valid	32.7K	—	495 ms	—	—	1	—	—
Prefill	32K	1	Valid	31.1K	—	1,036 ms	—	—	1	—	—
Prefill	64K	1	Valid	25.6K	—	2,510 ms	—	—	1	—	—
Prefill	128K	1	Valid	18.5K	—	6,931 ms	—	—	1	—	—
Prefill	251K	1	Valid	12.1K	—	20,748 ms	—	—	1	—	—
Prefill	256K	1	Valid	12K	—	21,323 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	185.7	186.2	66 ms	5.4 ms	1.0	1	5,570	30.0 s
Sustained decode	8K	2	Valid	228.9	113.9	87 ms	8.8 ms	2.9	0	6,866	30.0 s
Sustained decode	8K	4	Valid	611.4	154.0	97 ms	6.5 ms	4.0	4	18,340	30.0 s
Sustained decode	8K	8	Valid	962.7	128.7	132 ms	7.8 ms	10.3	8	28,880	30.0 s
Sustained decode	8K	16	Valid	1,447.3	91.1	201 ms	11.0 ms	24.0	0	43,408	30.0 s
Sustained decode	8K	32	Valid	2,619.0	79.9	341 ms	12.5 ms	36.3	0	78,576	30.0 s
Sustained decode	8K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	182.7	183.3	84 ms	5.5 ms	1.0	1	5,480	30.0 s
Sustained decode	16K	2	Valid	249.0	117.4	105 ms	8.5 ms	15.9	0	7,468	30.0 s
Sustained decode	16K	4	Valid	590.7	148.9	144 ms	6.7 ms	4.0	4	17,720	30.0 s
Sustained decode	16K	8	Valid	886.8	110.4	186 ms	9.1 ms	10.8	0	26,600	30.0 s
Sustained decode	16K	16	Valid	1,606.8	101.1	326 ms	9.9 ms	16.0	0	48,192	30.0 s
Sustained decode	16K	32	Valid	2,404.3	73.9	613 ms	13.5 ms	33.1	0	72,128	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	177.0	177.7	108 ms	5.6 ms	1.0	1	5,310	30.0 s
Sustained decode	32K	2	Valid	190.4	91.7	136 ms	10.9 ms	22.3	0	5,712	30.0 s
Sustained decode	32K	4	Valid	553.1	138.1	202 ms	7.2 ms	4.1	4	16,591	30.0 s
Sustained decode	32K	8	Valid	719.7	90.4	368 ms	11.1 ms	12.0	0	21,584	30.0 s
Sustained decode	32K	16	Valid	1,313.1	80.6	687 ms	12.4 ms	17.9	0	39,393	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	167.3	168.4	161 ms	5.9 ms	1.0	1	5,019	30.0 s
Sustained decode	64K	2	Valid	211.9	101.7	237 ms	9.8 ms	10.0	0	6,356	30.0 s
Sustained decode	64K	4	Valid	502.9	125.8	363 ms	8.0 ms	4.0	4	15,087	30.0 s
Sustained decode	64K	8	Valid	564.8	71.3	751 ms	14.0 ms	12.0	0	16,936	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	151.0	152.5	258 ms	6.6 ms	1.0	1	4,531	30.0 s
Sustained decode	128K	2	Valid	179.6	87.7	417 ms	11.4 ms	8.7	0	5,388	30.0 s
Sustained decode	128K	4	Valid	375.7	90.9	777 ms	11.0 ms	4.6	0	11,268	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	126.6	129.0	476 ms	7.8 ms	1.0	1	3,797	30.0 s
Sustained decode	251K	2	Valid	195.9	95.0	885 ms	10.5 ms	2.8	0	5,876	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	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.