Speed test matrix

Qwen3.5-9B-AWQ-4bit on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · W4A16 g326 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Nov 1, 2026, 10:15 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: W4A16 g32
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: MTP, 2 tokens lookahead
# 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 cyankiwi/Qwen3.5-9B-AWQ-4bit --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2}

Prefill

prompt tok/s by context length

8K context5,9831,361 ms TTFT

16K context5,7292,814 ms TTFT

32K context5,3755,969 ms TTFT

64K context4,72113,562 ms TTFT

128K context3,76333,987 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	178.9130 ms TTFT	352.9211 ms TTFT	614.8235 ms TTFT	973.0273 ms TTFT	1,500.0390 ms TTFT	1,873.0570 ms TTFT	2,437.7996 ms TTFT	2,713.11,519 ms TTFT
16K	126.9272 ms TTFT	237.5389 ms TTFT	440.0698 ms TTFT	570.51,091 ms TTFT	702.01,797 ms TTFT	796.83,300 ms TTFT	Over capacity	Over capacity
32K	95.0411 ms TTFT	178.3578 ms TTFT	315.4868 ms TTFT	387.61,515 ms TTFT	447.32,576 ms TTFT	501.14,635 ms TTFT	Over capacity	Over capacity
64K	61.2637 ms TTFT	120.6800 ms TTFT	205.51,158 ms TTFT	233.31,806 ms TTFT	262.03,016 ms TTFT	Over capacity	Over capacity	Over capacity
128K	35.91,023 ms TTFT	71.61,288 ms TTFT	127.21,889 ms TTFT	136.32,935 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	20.41,848 ms TTFT	39.22,397 ms TTFT	66.73,437 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	5,983.0	—	1,361 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,729.0	—	2,814 ms	—	—	1	—	—
Prefill	32K	1	Valid	5,375.0	—	5,969 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,721.0	—	13,562 ms	—	—	1	—	—
Prefill	128K	1	Valid	3,763.0	—	33,987 ms	—	—	1	—	—
Sustained decode	0	1	Valid	178.9	182.9	130 ms	5.5 ms	1.0	0	5,365	30.0 s
Sustained decode	0	2	Valid	352.9	181.2	211 ms	5.5 ms	2.0	0	10,585	30.0 s
Sustained decode	0	4	Valid	614.8	158.7	235 ms	6.3 ms	4.0	0	18,442	30.0 s
Sustained decode	0	8	Valid	973.0	126.5	273 ms	7.9 ms	8.0	0	29,187	30.0 s
Sustained decode	0	16	Valid	1,500.0	97.1	390 ms	10.3 ms	16.0	0	45,031	30.0 s
Sustained decode	0	32	Valid	1,873.0	60.1	570 ms	16.6 ms	32.0	0	56,213	30.0 s
Sustained decode	0	64	Valid	2,437.7	39.2	996 ms	25.5 ms	64.0	0	73,066	30.0 s
Sustained decode	0	128	Valid	2,713.1	21.8	1,519 ms	45.8 ms	128.0	0	81,434	30.0 s
Sustained decode	16K	1	Valid	126.9	128.2	272 ms	7.8 ms	1.0	0	3,805	30.0 s
Sustained decode	16K	2	Valid	237.5	120.9	389 ms	8.3 ms	2.0	0	7,124	30.0 s
Sustained decode	16K	4	Valid	440.0	110.4	698 ms	9.1 ms	4.0	0	13,193	30.0 s
Sustained decode	16K	8	Valid	570.5	71.9	1,091 ms	13.9 ms	8.0	0	17,117	30.0 s
Sustained decode	16K	16	Valid	702.0	43.6	1,797 ms	22.9 ms	16.0	0	21,059	30.0 s
Sustained decode	16K	32	Valid	796.8	24.4	3,300 ms	41.1 ms	32.0	0	23,901	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	95.0	96.4	411 ms	10.4 ms	1.0	0	2,849	30.0 s
Sustained decode	32K	2	Valid	178.3	90.7	578 ms	11.0 ms	2.0	0	5,349	30.0 s
Sustained decode	32K	4	Valid	315.4	79.0	868 ms	12.7 ms	4.0	0	9,456	30.0 s
Sustained decode	32K	8	Valid	387.6	48.5	1,515 ms	20.6 ms	8.0	0	11,613	30.0 s
Sustained decode	32K	16	Valid	447.3	28.2	2,576 ms	35.5 ms	16.0	0	13,413	30.0 s
Sustained decode	32K	32	Valid	501.1	15.1	4,635 ms	66.2 ms	32.0	0	14,982	29.9 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	61.2	62.4	637 ms	16.0 ms	1.0	0	1,835	30.0 s
Sustained decode	64K	2	Valid	120.6	60.9	800 ms	16.4 ms	2.0	0	3,614	30.0 s
Sustained decode	64K	4	Valid	205.5	52.1	1,158 ms	19.2 ms	4.0	0	6,163	30.0 s
Sustained decode	64K	8	Valid	233.3	29.8	1,806 ms	33.5 ms	8.0	0	6,982	29.9 s
Sustained decode	64K	16	Valid	262.0	16.4	3,016 ms	60.9 ms	16.0	0	7,839	29.9 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	35.9	36.7	1,023 ms	27.2 ms	1.0	0	1,074	30.0 s
Sustained decode	128K	2	Valid	71.6	36.5	1,288 ms	27.4 ms	2.0	0	2,147	30.0 s
Sustained decode	128K	4	Valid	127.2	32.1	1,889 ms	31.2 ms	4.0	0	3,818	30.0 s
Sustained decode	128K	8	Valid	136.3	17.3	2,935 ms	57.9 ms	8.0	0	4,082	29.9 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	20.4	21.1	1,848 ms	47.3 ms	1.0	0	610	29.9 s
Sustained decode	248K	2	Valid	39.2	20.1	2,397 ms	49.9 ms	2.0	0	1,174	30.0 s
Sustained decode	248K	4	Valid	66.7	17.4	3,437 ms	57.6 ms	4.0	0	2,000	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.