Speed test matrix

MiniMax-M3 on H200 SXM 141GB ×8

vllm · 0.29.0 · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 16, 2026, 19:33 UTC

Hardware

Type: Discrete GPU
GPU: H200 SXM
GPU count: 8
VRAM: 1128 GB (8 × 141 GB)
OS: linux

Engine

Engine: vllm
Version: 0.29.0
Quantization: BF16
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2

Run

Served model: MiniMaxAI/MiniMax-M3
Revision: main
Benchmark tool: llm-inference-bench v0.6.2
Decode test: Fixed duration
Prefill mode: Measured during decode warm-up
Time per measurement: 30 s
Max output: 4,096 tokens

Launch command

env FLASH_ATTENTION_CUTE_DSL_CACHE_ENABLED=1 VLLM_ENGINE_READY_TIMEOUT_S=3600 "$HOME/env_h200-glm53-vllm-v0290/bin/vllm" serve MiniMaxAI/MiniMax-M3 --served-model-name MiniMaxAI/MiniMax-M3 --trust-remote-code --tensor-parallel-size 8 --reasoning-parser minimax_m3 --enable-auto-tool-choice --tool-call-parser minimax_m3 --max-model-len 262144 --gpu-memory-utilization 0.84 --host 0.0.0.0 --port 8000 --api-key <redacted> --enforce-eager --block-size 128 --limit-mm-per-prompt '{"image":1,"video":0}'

Prefill

prompt tok/s by context length

8K context26.9K304 ms TTFT

16K context26.9K598 ms TTFT

32K context26.2K1,212 ms TTFT

64K context25.2K2,514 ms TTFT

128K context22.9K5,513 ms TTFT

251K context19.7K12,575 ms TTFT

256K context19.4K13,007 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	11.7129 ms TTFT	23.5231 ms TTFT	46.8334 ms TTFT	93.6365 ms TTFT	181.2425 ms TTFT	Over capacity	Over capacity	Over capacity
16K	11.7153 ms TTFT	23.3245 ms TTFT	46.5375 ms TTFT	93.3438 ms TTFT	180.8574 ms TTFT	Over capacity	Over capacity	Over capacity
32K	11.7198 ms TTFT	23.1291 ms TTFT	46.3428 ms TTFT	92.6606 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	11.7287 ms TTFT	23.2432 ms TTFT	46.3643 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	11.7508 ms TTFT	23.2686 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
251K	11.71,085 ms TTFT	Over capacity	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	26.9K	—	304 ms	—	—	1	—	—
Prefill	16K	1	Valid	26.9K	—	598 ms	—	—	1	—	—
Prefill	32K	1	Valid	26.2K	—	1,212 ms	—	—	1	—	—
Prefill	64K	1	Valid	25.2K	—	2,514 ms	—	—	1	—	—
Prefill	128K	1	Valid	22.9K	—	5,513 ms	—	—	1	—	—
Prefill	251K	1	Valid	19.7K	—	12,575 ms	—	—	1	—	—
Prefill	256K	1	Valid	19.4K	—	13,007 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	11.7	11.7	129 ms	85.6 ms	1.0	0	351	30.0 s
Sustained decode	8K	2	Valid	23.5	11.7	231 ms	85.3 ms	2.0	0	704	30.0 s
Sustained decode	8K	4	Valid	46.8	11.7	334 ms	85.5 ms	4.0	0	1,404	30.0 s
Sustained decode	8K	8	Valid	93.6	11.7	365 ms	85.4 ms	8.0	0	2,808	30.0 s
Sustained decode	8K	16	Valid	181.2	11.0	425 ms	90.9 ms	16.0	0	5,424	29.9 s
Sustained decode	8K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	8K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	11.7	11.7	153 ms	85.4 ms	1.0	0	351	29.9 s
Sustained decode	16K	2	Valid	23.3	11.7	245 ms	85.7 ms	2.0	0	700	30.0 s
Sustained decode	16K	4	Valid	46.5	11.6	375 ms	86.2 ms	4.0	0	1,392	29.9 s
Sustained decode	16K	8	Valid	93.3	11.6	438 ms	85.8 ms	8.0	0	2,792	29.9 s
Sustained decode	16K	16	Valid	180.8	11.3	574 ms	88.6 ms	16.0	0	5,424	30.0 s
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	11.7	11.7	198 ms	85.5 ms	1.0	0	350	29.9 s
Sustained decode	32K	2	Valid	23.1	11.5	291 ms	86.6 ms	2.0	0	692	29.9 s
Sustained decode	32K	4	Valid	46.3	11.6	428 ms	86.4 ms	4.0	0	1,388	30.0 s
Sustained decode	32K	8	Valid	92.6	11.5	606 ms	86.6 ms	8.0	0	2,776	30.0 s
Sustained decode	32K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	11.7	11.7	287 ms	85.5 ms	1.0	0	351	30.0 s
Sustained decode	64K	2	Valid	23.2	11.6	432 ms	86.2 ms	2.0	0	696	30.0 s
Sustained decode	64K	4	Valid	46.3	11.6	643 ms	86.5 ms	4.0	0	1,388	30.0 s
Sustained decode	64K	8	Over capacity	—	—	—	—	—	—	—	—
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	11.7	11.7	508 ms	85.3 ms	1.0	0	351	30.0 s
Sustained decode	128K	2	Valid	23.2	11.6	686 ms	86.2 ms	2.0	0	696	30.0 s
Sustained decode	128K	4	Over capacity	—	—	—	—	—	—	—	—
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	11.7	11.8	1,085 ms	85.0 ms	1.0	0	350	29.9 s
Sustained decode	251K	2	Over capacity	—	—	—	—	—	—	—	—
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.