Speed test matrix
gemma-4-12B-it-qat-w4a16-ct on CMP 170HX 64GB
vllm · 0.32.0rc1.dev249+g18305948a · W4A166 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
- 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 google/gemma-4-12B-it-qat-w4a16-ct --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only
Prefill
prompt tok/s by context length8K context3,4892,285 ms TTFT
16K context2,9955,266 ms TTFT
32K context2,41313,010 ms TTFT
64K context1,72736,271 ms TTFT
128K context1,102113,563 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 | 90.086 ms TTFT | 172.5143 ms TTFT | 322.4147 ms TTFT | 570.5177 ms TTFT | 889.7245 ms TTFT | 1,309.7376 ms TTFT | 1,751.6497 ms TTFT | 2,359.3819 ms TTFT |
| 16K | 70.9217 ms TTFT | 138.2277 ms TTFT | Looped | Looped | Looped | Looped | 1,216.22,589 ms TTFT | Over capacity |
| 32K | Looped | 123.9440 ms TTFT | Looped | Looped | Looped | 679.83,322 ms TTFT | Over capacity | Over capacity |
| 64K | Looped | Looped | 194.31,192 ms TTFT | Looped | Looped | Over capacity | Over capacity | Over capacity |
| 128K | 43.81,252 ms TTFT | 86.91,698 ms TTFT | Looped | 286.14,134 ms TTFT | Looped | Over capacity | Over capacity | Over capacity |
| 248K | 30.1Requests queued2,526 ms TTFT | 63.23,542 ms TTFT | 107.65,619 ms TTFT | 204.08,892 ms TTFT | 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. Looped: the model kept repeating the same text, so the benchmark discarded this measurement. 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 | 3,489.0 | — | 2,285 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 2,995.0 | — | 5,266 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 2,413.0 | — | 13,010 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 1,727.0 | — | 36,271 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 1,102.0 | — | 113,563 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 90.0 | 90.6 | 86 ms | 11.0 ms | 1.0 | 0 | 2,699 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 172.5 | 87.1 | 143 ms | 11.5 ms | 2.0 | 0 | 5,176 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 322.4 | 81.9 | 147 ms | 12.2 ms | 4.0 | 0 | 9,668 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 570.5 | 73.2 | 177 ms | 13.7 ms | 8.0 | 0 | 17,120 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 889.7 | 57.8 | 245 ms | 17.3 ms | 16.0 | 0 | 26,688 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 1,309.7 | 43.6 | 376 ms | 22.9 ms | 32.0 | 0 | 39,296 | 30.0 s |
| Sustained decode | 0 | 64 | Valid | 1,751.6 | 29.2 | 497 ms | 34.3 ms | 64.0 | 0 | 52,544 | 30.0 s |
| Sustained decode | 0 | 128 | Valid | 2,359.3 | 19.3 | 819 ms | 51.7 ms | 128.0 | 0 | 70,912 | 30.1 s |
| Sustained decode | 16K | 1 | Valid | 70.9 | 71.1 | 217 ms | 14.1 ms | 1.0 | 0 | 2,125 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 138.2 | 69.5 | 277 ms | 14.4 ms | 2.0 | 0 | 4,144 | 30.0 s |
| Sustained decode | 16K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 32 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 64 | Valid | 1,216.2 | 19.1 | 2,589 ms | 52.3 ms | 64.0 | 0 | 36,480 | 30.0 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 2 | Valid | 123.9 | 62.1 | 440 ms | 16.1 ms | 2.0 | 0 | 3,716 | 30.0 s |
| Sustained decode | 32K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 32 | Valid | 679.8 | 21.3 | 3,322 ms | 46.9 ms | 32.0 | 0 | 20,384 | 30.0 s |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 2 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 4 | Valid | 194.3 | 48.7 | 1,192 ms | 20.5 ms | 4.0 | 0 | 5,828 | 30.0 s |
| Sustained decode | 64K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 43.8 | 44.1 | 1,252 ms | 22.7 ms | 1.0 | 0 | 1,315 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 86.9 | 43.7 | 1,698 ms | 22.9 ms | 2.0 | 0 | 2,606 | 30.0 s |
| Sustained decode | 128K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 128K | 8 | Valid | 286.1 | 35.7 | 4,134 ms | 28.0 ms | 8.0 | 0 | 8,580 | 30.0 s |
| Sustained decode | 128K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 128K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 1 | Requests queued | 30.1 | 32.3 | 2,526 ms | 31.0 ms | 0.9 | 0 | 904 | 30.0 s |
| Sustained decode | 248K | 2 | Valid | 63.2 | 31.9 | 3,542 ms | 31.4 ms | 2.0 | 0 | 1,896 | 30.0 s |
| Sustained decode | 248K | 4 | Valid | 107.6 | 27.1 | 5,619 ms | 36.9 ms | 4.0 | 0 | 3,228 | 30.0 s |
| Sustained decode | 248K | 8 | Valid | 204.0 | 25.3 | 8,892 ms | 39.5 ms | 8.0 | 0 | 6,120 | 30.0 s |
| 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. Looped: the model kept repeating the same text, so the benchmark discarded this measurement. Over capacity: not run because this many requests at this context would exceed the server's KV cache.