Speed test matrix

Qwen3.8-27B-INT4 on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · INT4 W4A16 g1286 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 28, 2026, 19:33 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: INT4 W4A16 g128
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/Qwen3.8-27B-INT4 --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 context1,9494,203 ms TTFT

16K context1,8118,924 ms TTFT

32K context1,69318,978 ms TTFT

64K context1,47543,426 ms TTFT

128K context1,168109,491 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	53.3179 ms TTFT	103.7305 ms TTFT	197.8374 ms TTFT	370.3630 ms TTFT	607.61,164 ms TTFT	914.41,858 ms TTFT	1,132.73,030 ms TTFT	1,247.45,145 ms TTFT
16K	51.5453 ms TTFT	98.2604 ms TTFT	180.0963 ms TTFT	310.11,669 ms TTFT	460.82,997 ms TTFT	616.35,982 ms TTFT	Over capacity	Over capacity
32K	49.9738 ms TTFT	93.31,042 ms TTFT	165.21,617 ms TTFT	266.32,782 ms TTFT	371.05,002 ms TTFT	Over capacity	Over capacity	Over capacity
64K	47.31,579 ms TTFT	84.92,179 ms TTFT	141.63,409 ms TTFT	207.35,889 ms TTFT	267.210,784 ms TTFT	Over capacity	Over capacity	Over capacity
128K	42.62,020 ms TTFT	72.42,598 ms TTFT	110.43,920 ms TTFT	144.66,422 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	31.7Requests queued4,507 ms TTFT	56.45,847 ms TTFT	78.28,809 ms TTFT	Over capacity	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. 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	1,949.0	—	4,203 ms	—	—	1	—	—
Prefill	16K	1	Valid	1,811.0	—	8,924 ms	—	—	1	—	—
Prefill	32K	1	Valid	1,693.0	—	18,978 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,475.0	—	43,426 ms	—	—	1	—	—
Prefill	128K	1	Valid	1,168.0	—	109,491 ms	—	—	1	—	—
Sustained decode	0	1	Valid	53.3	53.4	179 ms	18.7 ms	1.0	0	1,600	30.0 s
Sustained decode	0	2	Valid	103.7	51.9	305 ms	19.3 ms	2.0	0	3,112	30.0 s
Sustained decode	0	4	Valid	197.8	49.6	374 ms	20.2 ms	4.0	0	5,932	30.0 s
Sustained decode	0	8	Valid	370.3	46.4	630 ms	21.5 ms	8.0	0	11,104	30.0 s
Sustained decode	0	16	Valid	607.6	38.1	1,164 ms	26.3 ms	16.0	0	18,224	30.0 s
Sustained decode	0	32	Valid	914.4	28.4	1,858 ms	35.3 ms	32.0	0	27,424	30.0 s
Sustained decode	0	64	Valid	1,132.7	17.2	3,030 ms	58.2 ms	64.0	0	33,984	30.0 s
Sustained decode	0	128	Valid	1,247.4	9.0	5,145 ms	111.2 ms	128.0	0	37,447	30.0 s
Sustained decode	16K	1	Valid	51.5	51.6	453 ms	19.4 ms	1.0	0	1,546	30.0 s
Sustained decode	16K	2	Valid	98.2	49.0	604 ms	20.4 ms	2.0	0	2,946	30.0 s
Sustained decode	16K	4	Valid	180.0	44.5	963 ms	22.5 ms	4.0	0	5,400	30.0 s
Sustained decode	16K	8	Valid	310.1	37.7	1,669 ms	26.5 ms	8.0	0	9,296	30.0 s
Sustained decode	16K	16	Valid	460.8	27.2	2,997 ms	36.8 ms	16.0	0	13,824	30.0 s
Sustained decode	16K	32	Valid	616.3	17.2	5,982 ms	58.3 ms	32.0	0	18,496	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	49.9	50.0	738 ms	20.0 ms	1.0	0	1,496	30.0 s
Sustained decode	32K	2	Valid	93.3	46.4	1,042 ms	21.5 ms	2.0	0	2,798	30.0 s
Sustained decode	32K	4	Valid	165.2	40.5	1,617 ms	24.7 ms	4.0	0	4,956	30.0 s
Sustained decode	32K	8	Valid	266.3	31.7	2,782 ms	31.5 ms	8.0	0	7,984	30.0 s
Sustained decode	32K	16	Valid	371.0	20.9	5,002 ms	47.9 ms	16.0	0	11,145	30.0 s
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	47.3	47.4	1,579 ms	21.1 ms	1.0	0	1,419	30.0 s
Sustained decode	64K	2	Valid	84.9	41.9	2,179 ms	23.9 ms	2.0	0	2,548	30.0 s
Sustained decode	64K	4	Valid	141.6	33.8	3,409 ms	29.6 ms	4.0	0	4,244	30.0 s
Sustained decode	64K	8	Valid	207.3	23.3	5,889 ms	43.0 ms	8.0	0	6,216	30.0 s
Sustained decode	64K	16	Valid	267.2	13.5	10,784 ms	74.3 ms	16.0	0	8,016	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	42.6	42.8	2,020 ms	23.4 ms	1.0	0	1,278	30.0 s
Sustained decode	128K	2	Valid	72.4	35.7	2,598 ms	28.0 ms	2.0	0	2,170	30.0 s
Sustained decode	128K	4	Valid	110.4	26.4	3,920 ms	37.9 ms	4.0	0	3,313	30.0 s
Sustained decode	128K	8	Valid	144.6	16.2	6,422 ms	61.7 ms	8.0	0	4,336	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	Requests queued	31.7	36.5	4,507 ms	27.4 ms	0.9	0	949	30.0 s
Sustained decode	248K	2	Valid	56.4	27.3	5,847 ms	36.7 ms	2.0	0	1,692	30.0 s
Sustained decode	248K	4	Valid	78.2	17.5	8,809 ms	57.2 ms	4.0	0	2,344	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	—	—	—	—	—	—	—	—

Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.