Speed test matrix
MiniMax-M2.7 on H200 SXM 141GB ×4
vllm · 0.29.0 · FP86 contexts (8K–187K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 187K, 192K
Hardware
- Type
- Discrete GPU
- GPU
- H200 SXM
- GPU count
- 4
- VRAM
- 564 GB (4 × 141 GB)
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.29.0
- Quantization
- FP8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
Run
- Served model
- nvidia/MiniMax-M2.7-DFlash
- 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-M2.7 --served-model-name nvidia/MiniMax-M2.7-DFlash --trust-remote-code --tensor-parallel-size 4 --reasoning-parser minimax_m2 --enable-auto-tool-choice --tool-call-parser minimax_m2 --max-model-len 196608 --gpu-memory-utilization 0.82 --host 0.0.0.0 --port 8000 --api-key <redacted> --kernel-config.enable_flashinfer_autotune=False --speculative-config '{"method":"dflash","model":"nvidia/MiniMax-M2.7-DFlash","num_speculative_tokens":8}'Prefill
prompt tok/s by context length8K context35.1K233 ms TTFT
16K context33.5K483 ms TTFT
32K context28.7K1,124 ms TTFT
64K context22K2,926 ms TTFT
128K context14.8K8,715 ms TTFT
187K context11.4K16,527 ms TTFT
192K context11.1K17,351 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 | 235.247 ms TTFT | 352.558 ms TTFT | 568.660 ms TTFT | 804.0131 ms TTFT | 1,264.3225 ms TTFT | Looped | Over capacity | Over capacity |
| 16K | 170.166 ms TTFT | 327.778 ms TTFT | 546.283 ms TTFT | 638.0199 ms TTFT | Looped | Over capacity | Over capacity | Over capacity |
| 32K | 183.3109 ms TTFT | 277.6114 ms TTFT | 520.1129 ms TTFT | 673.3357 ms TTFT | 1,057.3605 ms TTFT | Over capacity | Over capacity | Over capacity |
| 64K | 147.1199 ms TTFT | 245.6222 ms TTFT | 428.3329 ms TTFT | Looped | Over capacity | Over capacity | Over capacity | Over capacity |
| 128K | 155.7396 ms TTFT | 255.2408 ms TTFT | 355.9687 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 187K | 133.0639 ms TTFT | 202.5915 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
Looped: the model kept repeating the same text, so the benchmark discarded this measurement. 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 | 35.1K | — | 233 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 33.5K | — | 483 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 28.7K | — | 1,124 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 22K | — | 2,926 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 14.8K | — | 8,715 ms | — | — | 1 | — | — |
| Prefill | 187K | 1 | Valid | 11.4K | — | 16,527 ms | — | — | 1 | — | — |
| Prefill | 192K | 1 | Valid | 11.1K | — | 17,351 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 235.2 | 236.1 | 47 ms | 4.2 ms | 1.0 | 1 | 7,055 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 352.5 | 179.3 | 58 ms | 5.6 ms | 2.0 | 2 | 10,574 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 568.6 | 145.7 | 60 ms | 6.9 ms | 4.0 | 4 | 17,056 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 804.0 | 95.6 | 131 ms | 10.5 ms | 8.0 | 0 | 24,115 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 1,264.3 | 81.0 | 225 ms | 12.3 ms | 16.0 | 0 | 37,919 | 30.0 s |
| Sustained decode | 8K | 32 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 8K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 8K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 170.1 | 180.8 | 66 ms | 5.6 ms | 1.0 | 1 | 5,103 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 327.7 | 183.0 | 78 ms | 5.5 ms | 2.0 | 2 | 9,827 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 546.2 | 135.8 | 83 ms | 7.4 ms | 4.0 | 4 | 16,383 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 638.0 | 99.5 | 199 ms | 10.1 ms | 8.0 | 0 | 19,129 | 30.0 s |
| Sustained decode | 16K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 183.3 | 181.6 | 109 ms | 5.5 ms | 1.0 | 1 | 5,500 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 277.6 | 149.5 | 114 ms | 6.7 ms | 2.0 | 2 | 8,327 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 520.1 | 142.6 | 129 ms | 7.0 ms | 4.0 | 4 | 15,595 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 673.3 | 90.7 | 357 ms | 11.0 ms | 8.0 | 0 | 20,193 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 1,057.3 | 70.8 | 605 ms | 14.1 ms | 16.0 | 0 | 31,711 | 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 | 147.1 | 176.0 | 199 ms | 5.8 ms | 1.0 | 1 | 4,280 | 29.1 s |
| Sustained decode | 64K | 2 | Valid | 245.6 | 163.2 | 222 ms | 6.1 ms | 2.0 | 1 | 7,366 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 428.3 | 112.7 | 329 ms | 8.9 ms | 4.0 | 1 | 12,844 | 30.0 s |
| Sustained decode | 64K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| 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 | 155.7 | 158.4 | 396 ms | 6.3 ms | 1.0 | 1 | 4,670 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 255.2 | 124.5 | 408 ms | 8.0 ms | 2.0 | 2 | 7,651 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 355.9 | 96.6 | 687 ms | 10.4 ms | 4.0 | 1 | 10,669 | 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 | 187K | 1 | Valid | 133.0 | 147.4 | 639 ms | 6.8 ms | 1.0 | 1 | 3,989 | 30.0 s |
| Sustained decode | 187K | 2 | Valid | 202.5 | 103.7 | 915 ms | 9.8 ms | 2.0 | 0 | 6,073 | 30.0 s |
| Sustained decode | 187K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 187K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 187K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 187K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 187K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 187K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Looped: the model kept repeating the same text, so the benchmark discarded this measurement. Over capacity: not run because this many requests at this context would exceed the server's KV cache.