Speed test matrix
Qwen3.6-35B-A3B-FP8-dynamic on H200 NVL 141GB
vllm · 0.27.2rc1.dev113+g5cecfc013 · FP86 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K
Hardware
- Type
- Discrete GPU
- GPU
- H200 NVL
- VRAM
- 141 GB
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.27.2rc1.dev113+g5cecfc013
- Quantization
- FP8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
- Commit
5cecfc013750
Run
- Served model
- qwen
- 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-redhatai-vllm/bin/vllm" serve RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic --served-model-name qwen --trust-remote-code --tensor-parallel-size 1 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.881803 --host 0.0.0.0 --port 8000 --api-key <redacted>
Prefill
prompt tok/s by context length8K context40.3K203 ms TTFT
16K context40.2K403 ms TTFT
32K context41.4K780 ms TTFT
64K context39K1,652 ms TTFT
128K context29.7K4,336 ms TTFT
251K context20.2K12,477 ms TTFT
256K context19.9K12,936 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 | 168.8109 ms TTFT | 313.0181 ms TTFT | 568.1939 ms TTFT | 997.0275 ms TTFT | 1,624.4401 ms TTFT | 2,661.2586 ms TTFT | 4,153.1906 ms TTFT | 6,065.31,364 ms TTFT |
| 16K | 167.2107 ms TTFT | 309.3162 ms TTFT | 558.7224 ms TTFT | 962.2277 ms TTFT | 1,558.9397 ms TTFT | 2,502.7528 ms TTFT | 3,735.3806 ms TTFT | 5,222.51,442 ms TTFT |
| 32K | 163.2163 ms TTFT | 302.5249 ms TTFT | 539.4303 ms TTFT | 914.9392 ms TTFT | 1,446.9532 ms TTFT | 2,184.4848 ms TTFT | 3,092.41,395 ms TTFT | Over capacity |
| 64K | 156.7185 ms TTFT | 291.1187 ms TTFT | 503.3232 ms TTFT | 828.5444 ms TTFT | 1,246.9761 ms TTFT | 1,760.01,492 ms TTFT | Over capacity | Over capacity |
| 128K | 145.3333 ms TTFT | 268.2427 ms TTFT | 452.0606 ms TTFT | 705.0944 ms TTFT | 978.11,671 ms TTFT | Over capacity | Over capacity | Over capacity |
| 251K | 127.0620 ms TTFT | 240.0817 ms TTFT | 372.41,208 ms TTFT | 531.61,967 ms TTFT | 671.53,267 ms TTFT | 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 | 40.3K | — | 203 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 40.2K | — | 403 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 41.4K | — | 780 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 39K | — | 1,652 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 29.7K | — | 4,336 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 20.2K | — | 12,477 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 19.9K | — | 12,936 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 168.8 | 169.7 | 109 ms | 5.9 ms | 1.0 | 1 | 5,063 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 313.0 | 157.9 | 181 ms | 6.3 ms | 2.0 | 2 | 9,390 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 568.1 | 144.3 | 939 ms | 6.9 ms | 4.0 | 4 | 17,040 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 997.0 | 128.6 | 275 ms | 7.8 ms | 7.7 | 8 | 29,906 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 1,624.4 | 102.5 | 401 ms | 9.8 ms | 16.0 | 0 | 48,720 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 2,661.2 | 83.9 | 586 ms | 11.9 ms | 32.0 | 0 | 79,840 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 4,153.1 | 65.2 | 906 ms | 15.3 ms | 64.0 | 0 | 124,543 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 6,065.3 | 47.1 | 1,364 ms | 21.2 ms | 128.0 | 0 | 181,887 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 167.2 | 168.1 | 107 ms | 5.9 ms | 1.0 | 1 | 5,020 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 309.3 | 155.9 | 162 ms | 6.4 ms | 2.0 | 2 | 9,284 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 558.7 | 141.5 | 224 ms | 7.1 ms | 4.0 | 4 | 16,761 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 962.2 | 124.9 | 277 ms | 8.0 ms | 7.7 | 8 | 28,865 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,558.9 | 98.1 | 397 ms | 10.2 ms | 16.0 | 0 | 46,768 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 2,502.7 | 78.4 | 528 ms | 12.7 ms | 32.0 | 0 | 75,072 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 3,735.3 | 58.3 | 806 ms | 17.1 ms | 64.0 | 0 | 112,064 | 30.0 s |
| Sustained decode | 16K | 128 | Valid | 5,222.5 | 40.4 | 1,442 ms | 24.7 ms | 128.0 | 0 | 156,672 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 163.2 | 164.5 | 163 ms | 6.1 ms | 1.0 | 1 | 4,896 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 302.5 | 153.0 | 249 ms | 6.5 ms | 2.0 | 2 | 9,074 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 539.4 | 136.9 | 303 ms | 7.3 ms | 4.0 | 4 | 16,182 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 914.9 | 114.7 | 392 ms | 8.7 ms | 8.0 | 0 | 27,440 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 1,446.9 | 90.5 | 532 ms | 11.1 ms | 16.0 | 0 | 43,407 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 2,184.4 | 67.9 | 848 ms | 14.7 ms | 32.0 | 0 | 65,536 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 3,092.4 | 47.6 | 1,395 ms | 21.0 ms | 64.0 | 0 | 92,736 | 30.0 s |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 156.7 | 158.3 | 185 ms | 6.3 ms | 1.0 | 1 | 4,702 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 291.1 | 146.9 | 187 ms | 6.8 ms | 2.0 | 2 | 8,732 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 503.3 | 126.8 | 232 ms | 7.9 ms | 4.0 | 4 | 15,097 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 828.5 | 103.7 | 444 ms | 9.6 ms | 8.0 | 0 | 24,848 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 1,246.9 | 78.0 | 761 ms | 12.8 ms | 16.0 | 0 | 37,408 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 1,760.0 | 55.0 | 1,492 ms | 18.2 ms | 32.0 | 0 | 52,801 | 30.0 s |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 145.3 | 147.8 | 333 ms | 6.8 ms | 1.0 | 1 | 4,360 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 268.2 | 136.3 | 427 ms | 7.3 ms | 2.0 | 2 | 8,045 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 452.0 | 113.0 | 606 ms | 8.9 ms | 4.0 | 0 | 13,560 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 705.0 | 87.7 | 944 ms | 11.4 ms | 8.0 | 0 | 21,144 | 30.0 s |
| Sustained decode | 128K | 16 | Valid | 978.1 | 60.5 | 1,671 ms | 16.5 ms | 16.0 | 0 | 29,344 | 30.0 s |
| Sustained decode | 128K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 1 | Valid | 127.0 | 132.9 | 620 ms | 7.5 ms | 1.0 | 1 | 3,811 | 30.0 s |
| Sustained decode | 251K | 2 | Valid | 240.0 | 120.1 | 817 ms | 8.3 ms | 2.0 | 0 | 7,200 | 30.0 s |
| Sustained decode | 251K | 4 | Valid | 372.4 | 93.1 | 1,208 ms | 10.7 ms | 4.0 | 0 | 11,172 | 30.0 s |
| Sustained decode | 251K | 8 | Valid | 531.6 | 66.4 | 1,967 ms | 15.1 ms | 8.0 | 0 | 15,944 | 30.0 s |
| Sustained decode | 251K | 16 | Valid | 671.5 | 41.6 | 3,267 ms | 24.0 ms | 16.0 | 0 | 20,144 | 30.0 s |
| 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.