Speed test matrix
Qwen3.8-27B-INT8-W8A16-MTP on CMP 170HX 64GB
vllm · 0.32.0rc1.dev249+g18305948a · INT8 W8A166 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
- INT8 W8A16
- 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 lued/Qwen3.8-27B-INT8-W8A16-MTP --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 context1,6335,015 ms TTFT
16K context1,55210,410 ms TTFT
32K context1,47421,789 ms TTFT
64K context1,32148,514 ms TTFT
128K context1,091117,274 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 | 57.5230 ms TTFT | 110.7279 ms TTFT | 212.1494 ms TTFT | 347.9808 ms TTFT | 562.61,513 ms TTFT | 616.42,256 ms TTFT | Over capacity | Over capacity |
| 16K | 43.9791 ms TTFT | 83.61,362 ms TTFT | 127.62,618 ms TTFT | 189.04,397 ms TTFT | 231.77,162 ms TTFT | Over capacity | Over capacity | Over capacity |
| 32K | 37.0882 ms TTFT | 64.51,519 ms TTFT | 89.82,891 ms TTFT | 131.94,834 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 64K | 24.11,149 ms TTFT | 45.81,802 ms TTFT | 58.43,391 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 128K | 15.12,584 ms TTFT | 27.64,184 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 248K | 9.1Requests queued3,691 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 | 1,633.0 | — | 5,015 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 1,552.0 | — | 10,410 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 1,474.0 | — | 21,789 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 1,321.0 | — | 48,514 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 1,091.0 | — | 117,274 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 57.5 | 58.7 | 230 ms | 17.0 ms | 1.0 | 0 | 1,723 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 110.7 | 56.7 | 279 ms | 17.6 ms | 2.0 | 0 | 3,317 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 212.1 | 53.7 | 494 ms | 18.6 ms | 4.0 | 0 | 6,358 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 347.9 | 45.1 | 808 ms | 22.2 ms | 8.0 | 0 | 10,424 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 562.6 | 35.9 | 1,513 ms | 27.9 ms | 16.0 | 0 | 16,873 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 616.4 | 20.1 | 2,256 ms | 49.7 ms | 32.0 | 0 | 18,466 | 30.0 s |
| Sustained decode | 0 | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 43.9 | 44.0 | 791 ms | 22.7 ms | 1.0 | 0 | 1,315 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 83.6 | 42.4 | 1,362 ms | 23.6 ms | 2.0 | 0 | 2,507 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 127.6 | 32.2 | 2,618 ms | 31.1 ms | 4.0 | 0 | 3,825 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 189.0 | 23.4 | 4,397 ms | 42.8 ms | 8.0 | 0 | 5,657 | 29.9 s |
| Sustained decode | 16K | 16 | Valid | 231.7 | 14.1 | 7,162 ms | 70.7 ms | 16.0 | 0 | 6,933 | 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 | 37.0 | 37.4 | 882 ms | 26.7 ms | 1.0 | 0 | 1,111 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 64.5 | 32.8 | 1,519 ms | 30.5 ms | 2.0 | 0 | 1,934 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 89.8 | 22.7 | 2,891 ms | 44.0 ms | 4.0 | 0 | 2,687 | 29.9 s |
| Sustained decode | 32K | 8 | Valid | 131.9 | 16.6 | 4,834 ms | 60.2 ms | 8.0 | 0 | 3,956 | 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 | 24.1 | 24.8 | 1,149 ms | 40.4 ms | 1.0 | 0 | 721 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 45.8 | 23.7 | 1,802 ms | 42.2 ms | 2.0 | 0 | 1,375 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 58.4 | 15.1 | 3,391 ms | 66.1 ms | 4.0 | 0 | 1,749 | 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 | 15.1 | 15.7 | 2,584 ms | 63.5 ms | 1.0 | 0 | 452 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 27.6 | 14.4 | 4,184 ms | 69.3 ms | 2.0 | 0 | 824 | 29.9 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 | 9.1 | 10.1 | 3,691 ms | 98.6 ms | 0.9 | 0 | 271 | 29.8 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.