Speed test matrix

Qwen3.5-9B-FP8-dynamic on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP8 dynamic6 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 07:03 UTC

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 dynamic
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/Qwen3.5-9B-FP8-dynamic --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 length

8K context6,9021,180 ms TTFT

16K context6,6512,423 ms TTFT

32K context6,2135,164 ms TTFT

64K context5,41811,816 ms TTFT

128K context4,28729,833 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	93.1118 ms TTFT	180.1191 ms TTFT	346.9182 ms TTFT	650.7246 ms TTFT	1,181.0403 ms TTFT	1,793.0558 ms TTFT	2,673.1871 ms TTFT	3,499.41,230 ms TTFT
16K	88.8183 ms TTFT	166.6279 ms TTFT	304.1495 ms TTFT	518.3617 ms TTFT	812.31,090 ms TTFT	1,074.91,828 ms TTFT	Over capacity	Over capacity
32K	85.2319 ms TTFT	155.3415 ms TTFT	269.4534 ms TTFT	429.1840 ms TTFT	618.31,453 ms TTFT	765.12,777 ms TTFT	Over capacity	Over capacity
64K	79.1511 ms TTFT	136.4637 ms TTFT	219.5930 ms TTFT	319.01,439 ms TTFT	417.42,596 ms TTFT	Over capacity	Over capacity	Over capacity
128K	69.0824 ms TTFT	109.51,180 ms TTFT	159.81,707 ms TTFT	210.52,630 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	55.61,765 ms TTFT	79.32,270 ms TTFT	105.73,285 ms TTFT	Over capacity	Over capacity	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	6,902.0	—	1,180 ms	—	—	1	—	—
Prefill	16K	1	Valid	6,651.0	—	2,423 ms	—	—	1	—	—
Prefill	32K	1	Valid	6,213.0	—	5,164 ms	—	—	1	—	—
Prefill	64K	1	Valid	5,418.0	—	11,816 ms	—	—	1	—	—
Prefill	128K	1	Valid	4,287.0	—	29,833 ms	—	—	1	—	—
Sustained decode	0	1	Valid	93.1	93.2	118 ms	10.7 ms	1.0	0	2,791	30.0 s
Sustained decode	0	2	Valid	180.1	90.3	191 ms	11.1 ms	2.0	0	5,402	30.0 s
Sustained decode	0	4	Valid	346.9	87.0	182 ms	11.5 ms	4.0	0	10,408	30.0 s
Sustained decode	0	8	Valid	650.7	81.8	246 ms	12.2 ms	8.0	0	19,538	30.0 s
Sustained decode	0	16	Valid	1,181.0	74.4	403 ms	13.4 ms	16.0	0	35,424	30.0 s
Sustained decode	0	32	Valid	1,793.0	56.5	558 ms	17.7 ms	32.0	0	53,796	30.0 s
Sustained decode	0	64	Valid	2,673.1	42.0	871 ms	23.8 ms	64.0	0	80,192	30.0 s
Sustained decode	0	128	Valid	3,499.4	27.1	1,230 ms	36.8 ms	128.0	0	104,960	30.0 s
Sustained decode	16K	1	Valid	88.8	89.0	183 ms	11.2 ms	1.0	0	2,665	30.0 s
Sustained decode	16K	2	Valid	166.6	83.4	279 ms	12.0 ms	2.0	0	4,998	30.0 s
Sustained decode	16K	4	Valid	304.1	76.0	495 ms	13.2 ms	4.0	0	9,120	30.0 s
Sustained decode	16K	8	Valid	518.3	64.7	617 ms	15.5 ms	8.0	0	15,544	30.0 s
Sustained decode	16K	16	Valid	812.3	50.3	1,090 ms	19.9 ms	16.0	0	24,384	30.0 s
Sustained decode	16K	32	Valid	1,074.9	32.8	1,828 ms	30.5 ms	32.0	0	32,256	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	85.2	85.4	319 ms	11.7 ms	1.0	0	2,556	30.0 s
Sustained decode	32K	2	Valid	155.3	77.8	415 ms	12.9 ms	2.0	0	4,658	30.0 s
Sustained decode	32K	4	Valid	269.4	67.3	534 ms	14.9 ms	4.0	0	8,080	30.0 s
Sustained decode	32K	8	Valid	429.1	53.3	840 ms	18.8 ms	8.0	0	12,872	30.0 s
Sustained decode	32K	16	Valid	618.3	37.9	1,453 ms	26.4 ms	16.0	0	18,544	30.0 s
Sustained decode	32K	32	Valid	765.1	22.8	2,777 ms	43.9 ms	32.0	0	22,944	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	79.1	79.4	511 ms	12.6 ms	1.0	0	2,372	30.0 s
Sustained decode	64K	2	Valid	136.4	68.4	637 ms	14.6 ms	2.0	0	4,092	30.0 s
Sustained decode	64K	4	Valid	219.5	54.9	930 ms	18.2 ms	4.0	0	6,584	30.0 s
Sustained decode	64K	8	Valid	319.0	39.7	1,439 ms	25.2 ms	8.0	0	9,568	30.0 s
Sustained decode	64K	16	Valid	417.4	25.8	2,596 ms	38.8 ms	16.0	0	12,512	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	69.0	69.4	824 ms	14.4 ms	1.0	0	2,070	30.0 s
Sustained decode	128K	2	Valid	109.5	55.0	1,180 ms	18.2 ms	2.0	0	3,284	30.0 s
Sustained decode	128K	4	Valid	159.8	40.1	1,707 ms	25.0 ms	4.0	0	4,792	30.0 s
Sustained decode	128K	8	Valid	210.5	26.3	2,630 ms	38.1 ms	8.0	0	6,312	30.0 s
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	Valid	55.6	56.1	1,765 ms	17.8 ms	1.0	0	1,668	30.0 s
Sustained decode	248K	2	Valid	79.3	40.1	2,270 ms	25.0 ms	2.0	0	2,380	30.0 s
Sustained decode	248K	4	Valid	105.7	26.7	3,285 ms	37.5 ms	4.0	0	3,168	30.0 s
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	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.