Speed test matrix
Qwen3.5-27B-GPTQ-Int4 on H200 NVL 141GB
sglang · 0.5.6.post3.dev9160+g1cf2b8c54 · INT46 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
- sglang
- Version
- 0.5.6.post3.dev9160+g1cf2b8c54
- Quantization
- INT4
- Backend
- cuda
- Repository
- github.com/sgl-project/sglang
- Build
- cuda-13.2
- Commit
1cf2b8c54d81
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
env SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 "$HOME/env_h200-qwen-sglang/bin/sglang" serve --model-path Qwen/Qwen3.5-27B-GPTQ-Int4 --served-model-name qwen --trust-remote-code --tp 1 --reasoning-parser qwen3 --tool-call-parser qwen3_coder --context-length 262144 --mem-fraction-static 0.868286 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted> --mamba-radix-cache-strategy extra_buffer --speculative-algorithm EAGLE --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4
Prefill
prompt tok/s by context length8K context6,1451,326 ms TTFT
16K context6,1602,617 ms TTFT
32K context5,9995,351 ms TTFT
64K context5,57811,484 ms TTFT
128K context4,95125,846 ms TTFT
251K context4,05161,907 ms TTFT
256K context4,04063,317 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 | 208.594 ms TTFT | 348.9129 ms TTFT | 650.4867 ms TTFT | 999.7184 ms TTFT | 1,150.2349 ms TTFT | 1,910.6Requests queued1,883 ms TTFT | Over capacity | Over capacity |
| 16K | 193.3114 ms TTFT | 125.8221 ms TTFT | 437.0508 ms TTFT | 1,047.3259 ms TTFT | 1,057.8671 ms TTFT | 1,749.7Requests queued12,217 ms TTFT | Over capacity | Over capacity |
| 32K | 189.3166 ms TTFT | 145.9260 ms TTFT | 571.4342 ms TTFT | 709.3791 ms TTFT | 1,234.61,530 ms TTFT | Over capacity | Over capacity | Over capacity |
| 64K | 191.1308 ms TTFT | 225.8599 ms TTFT | 572.3848 ms TTFT | 537.61,823 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 128K | 170.3539 ms TTFT | 217.11,023 ms TTFT | 488.41,837 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 251K | 160.51,021 ms TTFT | 167.91,985 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
Requests queued: the server could not run all requests at once, so some waited in a queue. 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 | 6,145.0 | — | 1,326 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 6,160.0 | — | 2,617 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 5,999.0 | — | 5,351 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 5,578.0 | — | 11,484 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 4,951.0 | — | 25,846 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 4,051.0 | — | 61,907 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 4,040.0 | — | 63,317 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 208.5 | 207.8 | 94 ms | 4.8 ms | 1.0 | 1 | 6,253 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 348.9 | 177.1 | 129 ms | 5.7 ms | 2.3 | 2 | 10,463 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 650.4 | 164.8 | 867 ms | 6.1 ms | 4.6 | 4 | 19,510 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 999.7 | 127.6 | 184 ms | 7.8 ms | 10.1 | 8 | 29,986 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 1,150.2 | 72.9 | 349 ms | 13.7 ms | 24.0 | 0 | 34,491 | 30.0 s |
| Sustained decode | 8K | 32 | Requests queued | 1,910.6 | 62.0 | 1,883 ms | 16.1 ms | 30.5 | 26 | 57,286 | 30.0 s |
| Sustained decode | 8K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 8K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 193.3 | 194.0 | 114 ms | 5.2 ms | 1.0 | 1 | 5,798 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 125.8 | 63.9 | 221 ms | 15.6 ms | 29.0 | 0 | 3,768 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 437.0 | 105.1 | 508 ms | 9.5 ms | 15.1 | 0 | 13,107 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 1,047.3 | 132.5 | 259 ms | 7.5 ms | 7.9 | 8 | 31,420 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,057.8 | 67.2 | 671 ms | 14.9 ms | 24.0 | 0 | 31,698 | 30.0 s |
| Sustained decode | 16K | 32 | Requests queued | 1,749.7 | 57.2 | 12,217 ms | 17.5 ms | 30.6 | 15 | 52,485 | 30.0 s |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 189.3 | 190.6 | 166 ms | 5.2 ms | 1.0 | 1 | 5,677 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 145.9 | 73.1 | 260 ms | 13.7 ms | 17.9 | 0 | 4,376 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 571.4 | 147.0 | 342 ms | 6.8 ms | 5.7 | 4 | 17,134 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 709.3 | 89.0 | 791 ms | 11.2 ms | 11.9 | 0 | 21,262 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 1,234.6 | 76.2 | 1,530 ms | 13.1 ms | 18.1 | 0 | 37,000 | 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 | 191.1 | 190.6 | 308 ms | 5.3 ms | 1.0 | 1 | 5,733 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 225.8 | 107.7 | 599 ms | 9.3 ms | 10.7 | 0 | 6,773 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 572.3 | 148.8 | 848 ms | 6.7 ms | 3.9 | 4 | 17,160 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 537.6 | 70.3 | 1,823 ms | 14.2 ms | 12.0 | 0 | 16,126 | 30.0 s |
| 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 | 170.3 | 172.0 | 539 ms | 5.8 ms | 1.0 | 1 | 5,108 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 217.1 | 104.1 | 1,023 ms | 9.6 ms | 6.2 | 0 | 6,510 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 488.4 | 126.4 | 1,837 ms | 7.9 ms | 3.9 | 3 | 14,652 | 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 | 251K | 1 | Valid | 160.5 | 160.3 | 1,021 ms | 6.3 ms | 1.0 | 1 | 4,813 | 30.0 s |
| Sustained decode | 251K | 2 | Valid | 167.9 | 86.0 | 1,985 ms | 11.6 ms | 5.0 | 0 | 5,032 | 30.0 s |
| Sustained decode | 251K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| 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 | — | — | — | — | — | — | — | — |
Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.