Speed test matrix
Hy4-preview-FP8 on H200 SXM 141GB ×8
vllm · 0.29.0 · MXFP84 contexts (8K–59K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 59K, 64K
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
- MXFP8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
Run
- Served model
- tencent/Hy4-preview-FP8
- 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 VLLM_USE_V2_MODEL_RUNNER=1 FLASH_ATTENTION_CUTE_DSL_CACHE_ENABLED=1 VLLM_ENGINE_READY_TIMEOUT_S=3600 "$HOME/env_h200-glm53-vllm-v0290/bin/vllm" serve tencent/Hy4-preview-FP8 --served-model-name tencent/Hy4-preview-FP8 --trust-remote-code --tensor-parallel-size 8 --reasoning-parser hy_v4 --enable-auto-tool-choice --tool-call-parser hy_v4 --max-model-len 65536 --gpu-memory-utilization 0.88 --host 0.0.0.0 --port 8000 --api-key <redacted> --kernel-config.enable_flashinfer_autotune=False
Prefill
prompt tok/s by context length8K context4,6001,781 ms TTFT
16K context4,6143,512 ms TTFT
32K context4,5707,058 ms TTFT
59K context4,47413,263 ms TTFT
64K context4,46314,420 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 | 38.872 ms TTFT | 69.7105 ms TTFT | 131.6144 ms TTFT | 236.6163 ms TTFT | 425.4275 ms TTFT | Over capacity | Over capacity | Over capacity |
| 16K | 38.893 ms TTFT | 69.7172 ms TTFT | 131.4171 ms TTFT | 236.4239 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 32K | 38.8143 ms TTFT | 69.6190 ms TTFT | 131.2296 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 59K | 38.7205 ms TTFT | 69.4268 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 | 4,600.0 | — | 1,781 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 4,614.0 | — | 3,512 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 4,570.0 | — | 7,058 ms | — | — | 1 | — | — |
| Prefill | 59K | 1 | Valid | 4,474.0 | — | 13,263 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 4,463.0 | — | 14,420 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 38.8 | 38.8 | 72 ms | 25.8 ms | 1.0 | 0 | 1,164 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 69.7 | 34.9 | 105 ms | 28.7 ms | 2.0 | 0 | 2,092 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 131.6 | 33.0 | 144 ms | 30.3 ms | 4.0 | 0 | 3,944 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 236.6 | 29.6 | 163 ms | 33.8 ms | 8.0 | 0 | 7,096 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 425.4 | 26.6 | 275 ms | 37.5 ms | 16.0 | 0 | 12,752 | 30.0 s |
| Sustained decode | 8K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 8K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 8K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 38.8 | 38.8 | 93 ms | 25.8 ms | 1.0 | 0 | 1,164 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 69.7 | 34.8 | 172 ms | 28.7 ms | 2.0 | 0 | 2,090 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 131.4 | 32.9 | 171 ms | 30.4 ms | 4.0 | 0 | 3,940 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 236.4 | 29.6 | 239 ms | 33.8 ms | 8.0 | 0 | 7,088 | 30.0 s |
| Sustained decode | 16K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 38.8 | 38.8 | 143 ms | 25.8 ms | 1.0 | 0 | 1,163 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 69.6 | 34.8 | 190 ms | 28.7 ms | 2.0 | 0 | 2,088 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 131.2 | 32.8 | 296 ms | 30.5 ms | 4.0 | 0 | 3,936 | 30.0 s |
| Sustained decode | 32K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| 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 | 59K | 1 | Valid | 38.7 | 38.7 | 205 ms | 25.8 ms | 1.0 | 0 | 1,161 | 30.0 s |
| Sustained decode | 59K | 2 | Valid | 69.4 | 34.7 | 268 ms | 28.8 ms | 2.0 | 0 | 2,082 | 30.0 s |
| Sustained decode | 59K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 59K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 59K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 59K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 59K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 59K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.