Speed test matrix
Qwen3.6-35B-A3B-FP8 on RTX PRO 6000 Blackwell Server Edition 96GB
vllm · 0.30.1rc1.dev452+g73a583112 · FP86 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K
Hardware
- Type
- Discrete GPU
- GPU
- RTX PRO 6000 Blackwell Server Edition
- VRAM
- 96 GB
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.30.1rc1.dev452+g73a583112
- Quantization
- FP8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
- Commit
73a5831127a9
Run
- Served model
- qwen
- Revision
- main
- Benchmark tool
- llm-inference-bench v0.7.6
- Decode test
- Fixed duration
- Prefill mode
- Measured during decode warm-up
- Time per measurement
- 30 s
- Max output
- 4,096 tokens
Launch command
env MAX_JOBS=2 FLASHINFER_NVCC_THREADS=1 TORCHINDUCTOR_COMPILE_THREADS=1 "$HOME/env_rtxpro6k-qwen-vllm-73a5831127a9/bin/vllm" serve Qwen/Qwen3.6-35B-A3B-FP8 --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 0.83 --host 0.0.0.0 --port 8000 --api-key <redacted> --attention-backend FLASHINFER --linear-backend cutlass --moe-backend triton
Prefill
prompt tok/s by context length8K context37.6K218 ms TTFT
16K context36.7K442 ms TTFT
32K context34.4K940 ms TTFT
64K context27.6K2,338 ms TTFT
128K context18.7K6,873 ms TTFT
251K context11.6K21,687 ms TTFT
256K context11.4K22,468 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 | 204.075 ms TTFT | 313.8130 ms TTFT | 550.6162 ms TTFT | 848.9232 ms TTFT | 1,314.5335 ms TTFT | 1,955.4488 ms TTFT | 2,784.6906 ms TTFT | 3,746.31,540 ms TTFT |
| 16K | 199.868 ms TTFT | 310.697 ms TTFT | 544.9146 ms TTFT | 830.7240 ms TTFT | 1,289.4360 ms TTFT | 1,893.2517 ms TTFT | 2,641.1807 ms TTFT | Over capacity |
| 32K | 191.0107 ms TTFT | 300.1183 ms TTFT | 524.8241 ms TTFT | 798.8353 ms TTFT | 1,206.7541 ms TTFT | 1,731.8832 ms TTFT | Over capacity | Over capacity |
| 64K | 175.8123 ms TTFT | 280.3155 ms TTFT | 499.7142 ms TTFT | 760.6390 ms TTFT | 1,129.8674 ms TTFT | Over capacity | Over capacity | Over capacity |
| 128K | 149.9348 ms TTFT | 245.5463 ms TTFT | 449.5631 ms TTFT | 672.2966 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 251K | 118.5736 ms TTFT | 203.4897 ms TTFT | 373.11,367 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 | 37.6K | — | 218 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 36.7K | — | 442 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 34.4K | — | 940 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 27.6K | — | 2,338 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 18.7K | — | 6,873 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 11.6K | — | 21,687 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 11.4K | — | 22,468 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 204.0 | 204.6 | 75 ms | 4.9 ms | 1.0 | 1 | 6,119 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 313.8 | 158.0 | 130 ms | 6.3 ms | 2.0 | 2 | 9,414 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 550.6 | 140.5 | 162 ms | 7.1 ms | 4.0 | 4 | 16,517 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 848.9 | 106.8 | 232 ms | 9.4 ms | 8.0 | 0 | 25,464 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 1,314.5 | 83.1 | 335 ms | 12.0 ms | 16.0 | 0 | 39,424 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 1,955.4 | 61.9 | 488 ms | 16.2 ms | 32.0 | 0 | 58,656 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 2,784.6 | 43.9 | 906 ms | 22.8 ms | 64.0 | 0 | 83,540 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 3,746.3 | 29.1 | 1,540 ms | 34.4 ms | 128.0 | 0 | 112,384 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 199.8 | 200.4 | 68 ms | 5.0 ms | 1.0 | 1 | 5,994 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 310.6 | 156.1 | 97 ms | 6.4 ms | 2.0 | 2 | 9,318 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 544.9 | 138.0 | 146 ms | 7.2 ms | 4.0 | 4 | 16,345 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 830.7 | 104.5 | 240 ms | 9.6 ms | 8.0 | 0 | 24,921 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,289.4 | 81.4 | 360 ms | 12.3 ms | 16.0 | 0 | 38,672 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 1,893.2 | 59.8 | 517 ms | 16.7 ms | 32.0 | 0 | 56,768 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 2,641.1 | 41.6 | 807 ms | 24.1 ms | 64.0 | 0 | 79,232 | 30.0 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 191.0 | 191.8 | 107 ms | 5.2 ms | 1.0 | 1 | 5,730 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 300.1 | 151.3 | 183 ms | 6.6 ms | 2.0 | 2 | 9,002 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 524.8 | 133.6 | 241 ms | 7.5 ms | 4.0 | 4 | 15,744 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 798.8 | 100.4 | 353 ms | 10.0 ms | 8.0 | 0 | 23,960 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 1,206.7 | 75.9 | 541 ms | 13.2 ms | 16.0 | 0 | 36,192 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 1,731.8 | 54.1 | 832 ms | 18.5 ms | 32.0 | 0 | 51,966 | 30.0 s |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 175.8 | 176.8 | 123 ms | 5.7 ms | 1.0 | 1 | 5,275 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 280.3 | 141.7 | 155 ms | 7.1 ms | 2.0 | 2 | 8,409 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 499.7 | 125.7 | 142 ms | 8.0 ms | 4.0 | 4 | 14,990 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 760.6 | 95.2 | 390 ms | 10.5 ms | 8.0 | 0 | 22,816 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 1,129.8 | 70.6 | 674 ms | 14.2 ms | 16.0 | 0 | 33,888 | 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 | 149.9 | 152.1 | 348 ms | 6.6 ms | 1.0 | 1 | 4,497 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 245.5 | 124.7 | 463 ms | 8.0 ms | 2.0 | 2 | 7,365 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 449.5 | 111.9 | 631 ms | 8.9 ms | 4.0 | 0 | 13,484 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 672.2 | 83.0 | 966 ms | 12.0 ms | 8.0 | 0 | 20,177 | 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 | 118.5 | 118.7 | 736 ms | 8.4 ms | 1.0 | 0 | 3,555 | 30.0 s |
| Sustained decode | 251K | 2 | Valid | 203.4 | 101.5 | 897 ms | 9.9 ms | 2.0 | 0 | 6,101 | 30.0 s |
| Sustained decode | 251K | 4 | Valid | 373.1 | 92.5 | 1,367 ms | 10.8 ms | 4.0 | 0 | 11,192 | 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.