Speed test matrix
NVIDIA-Nemotron-Nano-9B-v2-quantized.w4a16 on CMP 170HX 64GB
vllm · 0.32.0rc1.dev249+g18305948a · W4A165 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K
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
- W4A16
- 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 RedHatAI/NVIDIA-Nemotron-Nano-9B-v2-quantized.w4a16 --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code
Prefill
prompt tok/s by context length8K context4,9211,695 ms TTFT
16K context4,7763,456 ms TTFT
32K context4,6647,044 ms TTFT
64K context4,40414,888 ms TTFT
120K context3,99430,743 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 | 138.891 ms TTFT | 259.1562 ms TTFT | 475.4163 ms TTFT | 827.5257 ms TTFT | 1,267.2464 ms TTFT | 1,666.0699 ms TTFT | 1,932.81,167 ms TTFT | 2,026.41,731 ms TTFT |
| 16K | 133.8259 ms TTFT | 246.4387 ms TTFT | 437.5567 ms TTFT | 720.1946 ms TTFT | 1,028.91,692 ms TTFT | 1,283.33,330 ms TTFT | 1,431.66,775 ms TTFT | Over capacity |
| 32K | 129.5246 ms TTFT | 234.7304 ms TTFT | 402.9492 ms TTFT | 637.0828 ms TTFT | 871.51,570 ms TTFT | 1,046.72,897 ms TTFT | 1,145.64,079 ms TTFT | Over capacity |
| 64K | 122.1678 ms TTFT | 212.4864 ms TTFT | 345.71,232 ms TTFT | 513.81,982 ms TTFT | 661.43,535 ms TTFT | 757.56,731 ms TTFT | Over capacity | Over capacity |
| 120K | 110.8864 ms TTFT | 181.51,087 ms TTFT | 278.51,693 ms TTFT | 385.12,667 ms TTFT | 465.44,827 ms TTFT | 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 | 4,921.0 | — | 1,695 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 4,776.0 | — | 3,456 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 4,664.0 | — | 7,044 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 4,404.0 | — | 14,888 ms | — | — | 1 | — | — |
| Prefill | 120K | 1 | Valid | 3,994.0 | — | 30,743 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 138.8 | 139.0 | 91 ms | 7.2 ms | 1.0 | 0 | 4,163 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 259.1 | 129.8 | 562 ms | 7.7 ms | 2.0 | 0 | 7,774 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 475.4 | 119.2 | 163 ms | 8.4 ms | 4.0 | 0 | 14,260 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 827.5 | 103.9 | 257 ms | 9.6 ms | 8.0 | 0 | 24,824 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 1,267.2 | 79.6 | 464 ms | 12.6 ms | 16.0 | 0 | 38,048 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 1,666.0 | 52.2 | 699 ms | 19.2 ms | 32.0 | 0 | 49,985 | 30.0 s |
| Sustained decode | 0 | 64 | Valid | 1,932.8 | 30.1 | 1,167 ms | 33.2 ms | 64.0 | 0 | 57,984 | 30.0 s |
| Sustained decode | 0 | 128 | Valid | 2,026.4 | 15.5 | 1,731 ms | 64.4 ms | 128.0 | 0 | 60,928 | 30.1 s |
| Sustained decode | 16K | 1 | Valid | 133.8 | 134.0 | 259 ms | 7.5 ms | 1.0 | 0 | 4,013 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 246.4 | 123.2 | 387 ms | 8.1 ms | 2.0 | 0 | 7,392 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 437.5 | 108.9 | 567 ms | 9.2 ms | 4.0 | 0 | 13,124 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 720.1 | 88.9 | 946 ms | 11.3 ms | 8.0 | 0 | 21,600 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,028.9 | 62.3 | 1,692 ms | 16.1 ms | 16.0 | 0 | 30,896 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 1,283.3 | 37.3 | 3,330 ms | 26.8 ms | 32.0 | 0 | 38,496 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 1,431.6 | 19.5 | 6,775 ms | 51.3 ms | 64.0 | 0 | 42,944 | 30.0 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 129.5 | 129.8 | 246 ms | 7.7 ms | 1.0 | 0 | 3,885 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 234.7 | 117.6 | 304 ms | 8.5 ms | 2.0 | 0 | 7,040 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 402.9 | 100.9 | 492 ms | 9.9 ms | 4.0 | 0 | 12,086 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 637.0 | 79.6 | 828 ms | 12.6 ms | 8.0 | 0 | 19,104 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 871.5 | 54.4 | 1,570 ms | 18.4 ms | 16.0 | 0 | 26,176 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 1,046.7 | 32.5 | 2,897 ms | 30.7 ms | 32.0 | 0 | 31,392 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 1,145.6 | 17.4 | 4,079 ms | 57.6 ms | 64.0 | 0 | 34,369 | 30.0 s |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 122.1 | 122.4 | 678 ms | 8.2 ms | 1.0 | 0 | 3,662 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 212.4 | 106.1 | 864 ms | 9.4 ms | 2.0 | 0 | 6,370 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 345.7 | 85.6 | 1,232 ms | 11.7 ms | 4.0 | 0 | 10,372 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 513.8 | 62.3 | 1,982 ms | 16.0 ms | 8.0 | 0 | 15,408 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 661.4 | 38.6 | 3,535 ms | 25.9 ms | 16.0 | 0 | 19,840 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 757.5 | 20.5 | 6,731 ms | 48.8 ms | 32.0 | 0 | 22,720 | 30.0 s |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 1 | Valid | 110.8 | 111.3 | 864 ms | 9.0 ms | 1.0 | 0 | 3,323 | 30.0 s |
| Sustained decode | 120K | 2 | Valid | 181.5 | 91.0 | 1,087 ms | 11.0 ms | 2.0 | 0 | 5,444 | 30.0 s |
| Sustained decode | 120K | 4 | Valid | 278.5 | 69.3 | 1,693 ms | 14.4 ms | 4.0 | 0 | 8,356 | 30.0 s |
| Sustained decode | 120K | 8 | Valid | 385.1 | 47.2 | 2,667 ms | 21.2 ms | 8.0 | 0 | 11,552 | 30.0 s |
| Sustained decode | 120K | 16 | Valid | 465.4 | 27.7 | 4,827 ms | 36.0 ms | 16.0 | 0 | 13,952 | 30.0 s |
| Sustained decode | 120K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.