Speed test matrix

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

vllm · 0.32.0rc1.dev249+g18305948a · AWQ3 contexts (0–32K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 05:08 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: AWQ
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-AWQ --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 context3,5232,329 ms TTFT

16K context3,1445,166 ms TTFT

32K context2,65112,200 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	94.760 ms TTFT	179.6103 ms TTFT	340.9118 ms TTFT	624.8145 ms TTFT	1,026.3206 ms TTFT	1,564.7315 ms TTFT	Over capacity	Over capacity
16K	77.3181 ms TTFT	132.5191 ms TTFT	212.6293 ms TTFT	305.0462 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	65.4285 ms TTFT	103.5384 ms TTFT	149.5581 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	3,523.0	—	2,329 ms	—	—	1	—	—
Prefill	16K	1	Valid	3,144.0	—	5,166 ms	—	—	1	—	—
Prefill	32K	1	Valid	2,651.0	—	12,200 ms	—	—	1	—	—
Sustained decode	0	1	Valid	94.7	95.3	60 ms	10.5 ms	1.0	0	2,841	30.0 s
Sustained decode	0	2	Valid	179.6	90.6	103 ms	11.0 ms	2.0	0	5,386	30.0 s
Sustained decode	0	4	Valid	340.9	86.6	118 ms	11.6 ms	4.0	0	10,227	30.0 s
Sustained decode	0	8	Valid	624.8	80.0	145 ms	12.5 ms	8.0	0	18,744	30.0 s
Sustained decode	0	16	Valid	1,026.3	66.4	206 ms	15.1 ms	16.0	0	30,784	30.0 s
Sustained decode	0	32	Valid	1,564.7	51.1	315 ms	19.6 ms	32.0	0	46,944	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	77.3	77.6	181 ms	12.9 ms	1.0	0	2,319	30.0 s
Sustained decode	16K	2	Valid	132.5	66.5	191 ms	15.0 ms	2.0	0	3,974	30.0 s
Sustained decode	16K	4	Valid	212.6	53.4	293 ms	18.7 ms	4.0	0	6,376	30.0 s
Sustained decode	16K	8	Valid	305.0	38.2	462 ms	26.2 ms	8.0	0	9,144	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	65.4	65.6	285 ms	15.2 ms	1.0	0	1,962	30.0 s
Sustained decode	32K	2	Valid	103.5	51.9	384 ms	19.3 ms	2.0	0	3,104	30.0 s
Sustained decode	32K	4	Valid	149.5	37.4	581 ms	26.7 ms	4.0	0	4,484	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.