Speed test matrix
Qwen3.8-27B-FP8 on CMP 170HX 64GB
vllm · 0.32.0rc1.dev249+g18305948a · FP86 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K
Hardware
- Type
- Discrete GPU
- GPU
- CMP 170HX
- VRAM
- 64 GB
- CPU
- 2x Intel Xeon E5-2630 v3
- RAM
- 64 GB
- OS
- Ubuntu 26.04.1
Engine
- Engine
- vllm
- Version
- 0.32.0rc1.dev249+g18305948a
- Quantization
- FP8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- ghcr.io/morrowmake/vllm-cmp170hx@sha256:5b45c4d6feafefe2eba962375b0a9a0b64fd22b9e05abb4d1966f14a32a2db79
Run
- Served model
- bench
- 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
- 8,192 tokens
Launch command
# Untuned out-of-the-box baseline. Spec decode: MTP, 2 tokens lookahead
# 1x NVIDIA CMP 170HX 64GB (cmpunlocker), 180 W limit, PCIe Gen2 x4 | 2x Intel Xeon E5-2630 v3 | 64 GB DDR4 | Ubuntu 26.04.1, NVIDIA driver 610.57.04 | vLLM: Morrowmake vllm-cmp170hx image
vllm serve Qwen/Qwen3.8-27B-FP8 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2}Prefill
prompt tok/s by context length8K context1,6884,851 ms TTFT
16K context1,59910,106 ms TTFT
32K context1,51821,161 ms TTFT
64K context1,35547,294 ms TTFT
128K context1,113114,896 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 | 8 | 16 | 32 | 64 | 128 |
|---|---|---|---|---|---|---|---|---|
| 0 | 66.7206 ms TTFT | 130.7390 ms TTFT | 230.0478 ms TTFT | 384.5781 ms TTFT | 594.91,392 ms TTFT | 652.02,215 ms TTFT | Over capacity | Over capacity |
| 16K | 48.1771 ms TTFT | 96.71,316 ms TTFT | 135.92,536 ms TTFT | 198.64,288 ms TTFT | 240.56,927 ms TTFT | Over capacity | Over capacity | Over capacity |
| 32K | 36.9884 ms TTFT | 69.91,505 ms TTFT | 93.72,803 ms TTFT | 135.24,694 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 64K | 26.51,179 ms TTFT | 49.61,833 ms TTFT | 59.43,372 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 128K | 16.42,532 ms TTFT | 30.24,193 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 248K | 10.4Requests queued3,750 ms TTFT | Over capacity | 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 | 1,688.0 | — | 4,851 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 1,599.0 | — | 10,106 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 1,518.0 | — | 21,161 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 1,355.0 | — | 47,294 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 1,113.0 | — | 114,896 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 66.7 | 67.9 | 206 ms | 14.7 ms | 1.0 | 0 | 2,000 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 130.7 | 66.4 | 390 ms | 15.1 ms | 2.0 | 0 | 3,919 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 230.0 | 58.5 | 478 ms | 17.1 ms | 4.0 | 0 | 6,895 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 384.5 | 49.2 | 781 ms | 20.3 ms | 8.0 | 0 | 11,532 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 594.9 | 38.2 | 1,392 ms | 26.2 ms | 16.0 | 0 | 17,823 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 652.0 | 20.9 | 2,215 ms | 47.8 ms | 32.0 | 0 | 19,542 | 30.0 s |
| Sustained decode | 0 | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 48.1 | 48.9 | 771 ms | 20.5 ms | 1.0 | 0 | 1,442 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 96.7 | 48.6 | 1,316 ms | 20.6 ms | 2.0 | 0 | 2,898 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 135.9 | 34.3 | 2,536 ms | 29.2 ms | 4.0 | 0 | 4,070 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 198.6 | 24.8 | 4,288 ms | 40.3 ms | 8.0 | 0 | 5,943 | 29.9 s |
| Sustained decode | 16K | 16 | Valid | 240.5 | 14.4 | 6,927 ms | 69.6 ms | 16.0 | 0 | 7,194 | 29.9 s |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 36.9 | 37.6 | 884 ms | 26.6 ms | 1.0 | 0 | 1,106 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 69.9 | 35.2 | 1,505 ms | 28.4 ms | 2.0 | 0 | 2,094 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 93.7 | 24.2 | 2,803 ms | 41.3 ms | 4.0 | 0 | 2,809 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 135.2 | 16.8 | 4,694 ms | 59.4 ms | 8.0 | 0 | 4,053 | 30.0 s |
| 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 | 64K | 1 | Valid | 26.5 | 27.0 | 1,179 ms | 37.0 ms | 1.0 | 0 | 793 | 29.9 s |
| Sustained decode | 64K | 2 | Valid | 49.6 | 25.3 | 1,833 ms | 39.6 ms | 2.0 | 0 | 1,485 | 29.9 s |
| Sustained decode | 64K | 4 | Valid | 59.4 | 15.5 | 3,372 ms | 64.4 ms | 4.0 | 0 | 1,781 | 30.0 s |
| Sustained decode | 64K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| 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 | 16.4 | 17.1 | 2,532 ms | 58.6 ms | 1.0 | 0 | 490 | 29.9 s |
| Sustained decode | 128K | 2 | Valid | 30.2 | 15.7 | 4,193 ms | 63.7 ms | 2.0 | 0 | 905 | 30.0 s |
| Sustained decode | 128K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| 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 | 248K | 1 | Requests queued | 10.4 | 11.6 | 3,750 ms | 85.8 ms | 0.9 | 0 | 312 | 29.9 s |
| Sustained decode | 248K | 2 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 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.