Speed test matrix

Qwen3-14B-FP8 on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP83 contexts (0–32K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 07:05 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
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 Qwen/Qwen3-14B-FP8 --max-model-len 40960 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context3,1222,628 ms TTFT

16K context2,8225,756 ms TTFT

32K context2,40613,441 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	74.764 ms TTFT	143.7107 ms TTFT	277.3124 ms TTFT	520.8131 ms TTFT	890.9209 ms TTFT	1,436.5301 ms TTFT	Over capacity	Over capacity
16K	63.4174 ms TTFT	110.8227 ms TTFT	183.7304 ms TTFT	273.4462 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	55.1261 ms TTFT	89.4393 ms TTFT	133.6572 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	3,122.0	—	2,628 ms	—	—	1	—	—
Prefill	16K	1	Valid	2,822.0	—	5,756 ms	—	—	1	—	—
Prefill	32K	1	Valid	2,406.0	—	13,441 ms	—	—	1	—	—
Sustained decode	0	1	Valid	74.7	75.1	64 ms	13.3 ms	1.0	0	2,242	30.0 s
Sustained decode	0	2	Valid	143.7	72.3	107 ms	13.8 ms	2.0	0	4,312	30.0 s
Sustained decode	0	4	Valid	277.3	70.0	124 ms	14.3 ms	4.0	0	8,316	30.0 s
Sustained decode	0	8	Valid	520.8	66.2	131 ms	15.1 ms	8.0	0	15,624	30.0 s
Sustained decode	0	16	Valid	890.9	57.1	209 ms	17.5 ms	16.0	0	26,720	30.0 s
Sustained decode	0	32	Valid	1,436.5	46.4	301 ms	21.6 ms	32.0	0	43,104	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	63.4	63.6	174 ms	15.7 ms	1.0	0	1,902	30.0 s
Sustained decode	16K	2	Valid	110.8	55.6	227 ms	18.0 ms	2.0	0	3,324	30.0 s
Sustained decode	16K	4	Valid	183.7	46.1	304 ms	21.7 ms	4.0	0	5,508	30.0 s
Sustained decode	16K	8	Valid	273.4	34.2	462 ms	29.2 ms	8.0	0	8,200	30.0 s
Sustained decode	16K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	55.1	55.2	261 ms	18.1 ms	1.0	0	1,652	30.0 s
Sustained decode	32K	2	Valid	89.4	44.8	393 ms	22.3 ms	2.0	0	2,682	30.0 s
Sustained decode	32K	4	Valid	133.6	33.4	572 ms	29.9 ms	4.0	0	4,004	30.0 s
Sustained decode	32K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—

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