Speed test matrix

Ministral-3-3B-Reasoning-2512 on H200 NVL 141GB

sglang · 0.5.6.post3.dev9335+gdd15fb57b · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 19, 2026, 15:14 UTC

Hardware

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

Engine

Engine: sglang
Version: 0.5.6.post3.dev9335+gdd15fb57b
Quantization: BF16
Backend: cuda
Repository: github.com/sgl-project/sglang
Build: cuda-13.2
Commit: dd15fb57b5ef

Run

Served model: mistral
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-mistralai-sglang/bin/sglang" serve --model-path mistralai/Ministral-3-3B-Reasoning-2512 --served-model-name mistral --trust-remote-code --tp 1 --reasoning-parser mistral --tool-call-parser mistral --context-length 262144 --mem-fraction-static 0.860692 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>

Prefill

prompt tok/s by context length

8K context50.1K164 ms TTFT

16K context47.5K342 ms TTFT

32K context38K852 ms TTFT

64K context25.9K2,493 ms TTFT

128K context15.5K8,292 ms TTFT

251K context8,80828,673 ms TTFT

256K context8,74929,442 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	265.052 ms TTFT	507.671 ms TTFT	932.581 ms TTFT	1,613.2118 ms TTFT	2,456.0175 ms TTFT	2,518.8437 ms TTFT	3,367.8875 ms TTFT	Over capacity
16K	249.570 ms TTFT	102.6102 ms TTFT	715.7143 ms TTFT	1,166.9199 ms TTFT	1,232.4425 ms TTFT	1,947.6815 ms TTFT	Over capacity	Over capacity
32K	226.092 ms TTFT	101.9155 ms TTFT	547.7219 ms TTFT	645.3408 ms TTFT	956.4770 ms TTFT	Over capacity	Over capacity	Over capacity
64K	190.2136 ms TTFT	93.5254 ms TTFT	265.2407 ms TTFT	385.9767 ms TTFT	436.81,478 ms TTFT	Over capacity	Over capacity	Over capacity
128K	144.1225 ms TTFT	51.2432 ms TTFT	92.3770 ms TTFT	140.51,497 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	99.6401 ms TTFT	26.6783 ms TTFT	53.11,453 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	50.1K	—	164 ms	—	—	1	—	—
Prefill	16K	1	Valid	47.5K	—	342 ms	—	—	1	—	—
Prefill	32K	1	Valid	38K	—	852 ms	—	—	1	—	—
Prefill	64K	1	Valid	25.9K	—	2,493 ms	—	—	1	—	—
Prefill	128K	1	Valid	15.5K	—	8,292 ms	—	—	1	—	—
Prefill	251K	1	Valid	8,808.0	—	28,673 ms	—	—	1	—	—
Prefill	256K	1	Valid	8,749.0	—	29,442 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	265.0	266.0	52 ms	3.8 ms	1.0	2	7,950	30.0 s
Sustained decode	8K	2	Valid	507.6	235.9	71 ms	4.3 ms	2.0	2	15,238	30.0 s
Sustained decode	8K	4	Valid	932.5	234.8	81 ms	4.3 ms	4.0	4	27,977	30.0 s
Sustained decode	8K	8	Valid	1,613.2	204.4	118 ms	4.9 ms	8.0	8	48,398	30.0 s
Sustained decode	8K	16	Valid	2,456.0	157.4	175 ms	6.4 ms	16.5	16	73,675	30.0 s
Sustained decode	8K	32	Valid	2,518.8	80.4	437 ms	12.4 ms	47.5	0	75,552	30.0 s
Sustained decode	8K	64	Valid	3,367.8	52.5	875 ms	19.0 ms	84.3	0	101,037	30.0 s
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	249.5	251.0	70 ms	4.0 ms	1.0	2	7,486	30.0 s
Sustained decode	16K	2	Valid	102.6	51.9	102 ms	19.3 ms	66.0	0	3,076	30.0 s
Sustained decode	16K	4	Valid	715.7	180.8	143 ms	5.7 ms	7.1	4	21,469	30.0 s
Sustained decode	16K	8	Valid	1,166.9	156.6	199 ms	6.5 ms	9.7	8	35,006	30.0 s
Sustained decode	16K	16	Valid	1,232.4	78.0	425 ms	12.8 ms	24.0	0	36,960	30.0 s
Sustained decode	16K	32	Valid	1,947.6	60.6	815 ms	16.5 ms	41.6	0	58,432	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	226.0	226.9	92 ms	4.4 ms	1.0	1	6,779	30.0 s
Sustained decode	32K	2	Valid	101.9	51.3	155 ms	19.5 ms	34.0	0	3,057	30.0 s
Sustained decode	32K	4	Valid	547.7	151.5	219 ms	6.7 ms	5.0	4	16,430	30.0 s
Sustained decode	32K	8	Valid	645.3	81.3	408 ms	12.3 ms	12.0	0	19,352	30.0 s
Sustained decode	32K	16	Valid	956.4	58.6	770 ms	17.1 ms	20.9	0	28,688	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	190.2	191.3	136 ms	5.2 ms	1.0	1	5,706	30.0 s
Sustained decode	64K	2	Valid	93.5	46.9	254 ms	21.3 ms	18.0	0	2,802	30.0 s
Sustained decode	64K	4	Valid	265.2	63.5	407 ms	15.7 ms	6.9	0	7,954	30.0 s
Sustained decode	64K	8	Valid	385.9	48.7	767 ms	20.6 ms	12.0	0	11,568	30.0 s
Sustained decode	64K	16	Valid	436.8	27.8	1,478 ms	35.9 ms	24.0	0	13,104	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	144.1	145.6	225 ms	6.9 ms	1.0	1	4,322	30.0 s
Sustained decode	128K	2	Valid	51.2	25.6	432 ms	39.1 ms	26.0	0	1,534	30.0 s
Sustained decode	128K	4	Valid	92.3	22.6	770 ms	44.2 ms	25.2	0	2,768	30.0 s
Sustained decode	128K	8	Valid	140.5	17.8	1,497 ms	56.1 ms	30.0	0	4,208	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	99.6	99.6	401 ms	10.0 ms	1.0	0	2,989	30.0 s
Sustained decode	251K	2	Valid	26.6	13.4	783 ms	74.8 ms	32.0	0	798	30.0 s
Sustained decode	251K	4	Valid	53.1	13.2	1,453 ms	75.9 ms	24.3	0	1,592	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.