Speed test matrix
Qwen3.5-27B on RTX PRO 6000 Blackwell Server Edition 96GB ×2
vllm · 0.30.1rc1.dev452+g73a583112 · BF166 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
- GPU count
- 2
- VRAM
- 192 GB (2 × 96 GB)
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.30.1rc1.dev452+g73a583112
- Quantization
- BF16
- 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 v0.8.0
- 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_rtxpro6k-qwen-vllm-73a5831127a9/bin/vllm" serve Qwen/Qwen3.5-27B --served-model-name qwen --trust-remote-code --tensor-parallel-size 2 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.91 --host 0.0.0.0 --port 8000 --api-key <redacted> --attention-backend FLASH_ATTN --speculative-config '{"method":"qwen3_next_mtp","num_speculative_tokens":2}'Prefill
prompt tok/s by context length8K context8,377977 ms TTFT
16K context8,4031,930 ms TTFT
32K context8,1143,980 ms TTFT
64K context7,5058,583 ms TTFT
128K context6,51819,743 ms TTFT
251K context5,17048,781 ms TTFT
256K context5,13950,046 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 | 76.0151 ms TTFT | 153.0237 ms TTFT | 313.4543 ms TTFT | 577.31,546 ms TTFT | 930.41,111 ms TTFT | 1,424.52,068 ms TTFT | 1,933.83,956 ms TTFT | 2,208.67,754 ms TTFT |
| 16K | 66.7174 ms TTFT | 134.3285 ms TTFT | 268.4599 ms TTFT | 505.7970 ms TTFT | 753.91,261 ms TTFT | 1,123.62,276 ms TTFT | 1,489.04,704 ms TTFT | Over capacity |
| 32K | 52.9228 ms TTFT | 104.2357 ms TTFT | 212.4735 ms TTFT | 403.9831 ms TTFT | 548.61,966 ms TTFT | 790.13,022 ms TTFT | Over capacity | Over capacity |
| 64K | 36.9333 ms TTFT | 74.1504 ms TTFT | 149.4830 ms TTFT | 282.81,380 ms TTFT | 352.62,330 ms TTFT | Over capacity | Over capacity | Over capacity |
| 128K | 23.0548 ms TTFT | 47.3796 ms TTFT | 92.81,198 ms TTFT | 184.31,908 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 251K | 14.1904 ms TTFT | 26.01,342 ms TTFT | 55.02,067 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 | 8,377.0 | — | 977 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 8,403.0 | — | 1,930 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 8,114.0 | — | 3,980 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 7,505.0 | — | 8,583 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 6,518.0 | — | 19,743 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 5,170.0 | — | 48,781 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 5,139.0 | — | 50,046 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 76.0 | 77.3 | 151 ms | 12.9 ms | 1.0 | 0 | 2,279 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 153.0 | 75.1 | 237 ms | 13.3 ms | 2.0 | 0 | 4,586 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 313.4 | 77.1 | 543 ms | 13.0 ms | 4.0 | 0 | 9,395 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 577.3 | 70.7 | 1,546 ms | 14.1 ms | 8.0 | 0 | 17,316 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 930.4 | 56.2 | 1,111 ms | 17.8 ms | 16.0 | 0 | 27,906 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 1,424.5 | 43.5 | 2,068 ms | 23.0 ms | 32.0 | 0 | 42,691 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 1,933.8 | 28.3 | 3,956 ms | 35.3 ms | 64.0 | 0 | 57,998 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 2,208.6 | 14.9 | 7,754 ms | 67.1 ms | 128.0 | 0 | 66,123 | 29.9 s |
| Sustained decode | 16K | 1 | Valid | 66.7 | 67.5 | 174 ms | 14.8 ms | 1.0 | 0 | 1,999 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 134.3 | 67.6 | 285 ms | 14.8 ms | 2.0 | 0 | 4,028 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 268.4 | 67.6 | 599 ms | 14.8 ms | 4.0 | 0 | 8,044 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 505.7 | 63.3 | 970 ms | 15.8 ms | 8.0 | 0 | 15,163 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 753.9 | 46.6 | 1,261 ms | 21.5 ms | 16.0 | 0 | 22,584 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 1,123.6 | 33.9 | 2,276 ms | 29.5 ms | 32.0 | 0 | 33,650 | 29.9 s |
| Sustained decode | 16K | 64 | Valid | 1,489.0 | 21.8 | 4,704 ms | 45.9 ms | 64.0 | 0 | 44,549 | 29.9 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 52.9 | 53.6 | 228 ms | 18.7 ms | 1.0 | 0 | 1,586 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 104.2 | 52.6 | 357 ms | 19.0 ms | 2.0 | 0 | 3,126 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 212.4 | 53.6 | 735 ms | 18.6 ms | 4.0 | 0 | 6,364 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 403.9 | 50.2 | 831 ms | 19.9 ms | 8.0 | 0 | 12,100 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 548.6 | 34.1 | 1,966 ms | 29.3 ms | 16.0 | 0 | 16,450 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 790.1 | 23.8 | 3,022 ms | 42.0 ms | 32.0 | 0 | 23,630 | 29.9 s |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 36.9 | 37.5 | 333 ms | 26.7 ms | 1.0 | 0 | 1,105 | 29.9 s |
| Sustained decode | 64K | 2 | Valid | 74.1 | 37.4 | 504 ms | 26.8 ms | 2.0 | 0 | 2,220 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 149.4 | 37.6 | 830 ms | 26.6 ms | 4.0 | 0 | 4,479 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 282.8 | 35.1 | 1,380 ms | 28.5 ms | 8.0 | 0 | 8,474 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 352.6 | 21.4 | 2,330 ms | 46.7 ms | 16.0 | 0 | 10,562 | 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 | 23.0 | 23.4 | 548 ms | 42.8 ms | 1.0 | 0 | 689 | 29.9 s |
| Sustained decode | 128K | 2 | Valid | 47.3 | 23.8 | 796 ms | 42.1 ms | 2.0 | 0 | 1,415 | 29.9 s |
| Sustained decode | 128K | 4 | Valid | 92.8 | 23.3 | 1,198 ms | 42.9 ms | 4.0 | 0 | 2,778 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 184.3 | 22.9 | 1,908 ms | 43.7 ms | 8.0 | 0 | 5,511 | 29.9 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 | 14.1 | 14.3 | 904 ms | 70.0 ms | 1.0 | 0 | 422 | 29.9 s |
| Sustained decode | 251K | 2 | Valid | 26.0 | 13.3 | 1,342 ms | 75.3 ms | 2.0 | 0 | 776 | 29.9 s |
| Sustained decode | 251K | 4 | Valid | 55.0 | 13.8 | 2,067 ms | 72.2 ms | 4.0 | 0 | 1,644 | 29.9 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.