Speed test matrix
internlm_internlm3-8b-instruct-autoround-int8-gs128-sym on CMP 170HX 64GB
vllm · 0.32.0rc1.dev249+g18305948a · AutoRound INT8 W8A16 gs1283 contexts (0–24K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 24K
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
- AutoRound INT8 W8A16 gs128
- 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: none
# 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 fbaldassarri/internlm_internlm3-8b-instruct-autoround-int8-gs128-sym --max-model-len 32768 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --hf-overrides {"quantization_config":{"quant_method":"gptq","bits":8,"group_size":128,"sym":true,"desc_act":false,"lm_head":false}}Prefill
prompt tok/s by context length8K context4,8782,045 ms TTFT
16K context4,1474,760 ms TTFT
24K context3,6118,181 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 | 95.268 ms TTFT | 189.5109 ms TTFT | 374.396 ms TTFT | 723.7136 ms TTFT | 1,339.5184 ms TTFT | 2,224.0261 ms TTFT | 3,134.2392 ms TTFT | 4,081.9583 ms TTFT |
| 16K | 87.2150 ms TTFT | 166.8285 ms TTFT | 303.5372 ms TTFT | 505.3590 ms TTFT | 744.21,016 ms TTFT | 959.91,840 ms TTFT | Over capacity | Over capacity |
| 24K | 84.0201 ms TTFT | 157.9345 ms TTFT | 275.2503 ms TTFT | 435.5782 ms TTFT | 602.51,391 ms TTFT | 741.42,326 ms TTFT | 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 | 4,878.0 | — | 2,045 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 4,147.0 | — | 4,760 ms | — | — | 1 | — | — |
| Prefill | 24K | 1 | Valid | 3,611.0 | — | 8,181 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 95.2 | 95.3 | 68 ms | 10.5 ms | 1.0 | 0 | 2,855 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 189.5 | 95.1 | 109 ms | 10.5 ms | 2.0 | 0 | 5,684 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 374.3 | 94.2 | 96 ms | 10.6 ms | 4.0 | 0 | 11,228 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 723.7 | 91.4 | 136 ms | 10.9 ms | 8.0 | 0 | 21,720 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 1,339.5 | 85.2 | 184 ms | 11.7 ms | 16.0 | 0 | 40,192 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 2,224.0 | 71.1 | 261 ms | 14.1 ms | 32.0 | 0 | 66,720 | 30.0 s |
| Sustained decode | 0 | 64 | Valid | 3,134.2 | 50.2 | 392 ms | 19.9 ms | 64.0 | 0 | 94,016 | 30.0 s |
| Sustained decode | 0 | 128 | Valid | 4,081.9 | 32.5 | 583 ms | 30.8 ms | 128.0 | 0 | 122,496 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 87.2 | 87.3 | 150 ms | 11.5 ms | 1.0 | 0 | 2,616 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 166.8 | 83.6 | 285 ms | 12.0 ms | 2.0 | 0 | 5,002 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 303.5 | 76.0 | 372 ms | 13.2 ms | 4.0 | 0 | 9,104 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 505.3 | 63.1 | 590 ms | 15.8 ms | 8.0 | 0 | 15,152 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 744.2 | 46.0 | 1,016 ms | 21.7 ms | 16.0 | 0 | 22,336 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 959.9 | 29.4 | 1,840 ms | 34.0 ms | 32.0 | 0 | 28,800 | 30.0 s |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 24K | 1 | Valid | 84.0 | 84.1 | 201 ms | 11.9 ms | 1.0 | 0 | 2,520 | 30.0 s |
| Sustained decode | 24K | 2 | Valid | 157.9 | 79.0 | 345 ms | 12.7 ms | 2.0 | 0 | 4,738 | 30.0 s |
| Sustained decode | 24K | 4 | Valid | 275.2 | 68.8 | 503 ms | 14.5 ms | 4.0 | 0 | 8,252 | 30.0 s |
| Sustained decode | 24K | 8 | Valid | 435.5 | 54.1 | 782 ms | 18.5 ms | 8.0 | 0 | 13,064 | 30.0 s |
| Sustained decode | 24K | 16 | Valid | 602.5 | 37.0 | 1,391 ms | 27.0 ms | 16.0 | 0 | 18,064 | 30.0 s |
| Sustained decode | 24K | 32 | Valid | 741.4 | 22.1 | 2,326 ms | 45.2 ms | 32.0 | 0 | 22,272 | 30.0 s |
| Sustained decode | 24K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 24K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.