Speed test matrix

Qwen3.6-35B-A3B-FP8 on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP86 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, 04:00 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/Qwen3.6-35B-A3B-FP8 --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 context6,5781,238 ms TTFT

16K context6,5562,458 ms TTFT

32K context6,5314,912 ms TTFT

64K context5,62111,390 ms TTFT

128K context4,25530,058 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	153.1131 ms TTFT	265.8230 ms TTFT	445.7245 ms TTFT	721.0352 ms TTFT	1,103.6407 ms TTFT	1,603.9639 ms TTFT	2,265.2939 ms TTFT	2,936.71,250 ms TTFT
16K	147.4291 ms TTFT	247.1417 ms TTFT	403.6618 ms TTFT	603.2922 ms TTFT	854.81,709 ms TTFT	1,157.53,082 ms TTFT	Over capacity	Over capacity
32K	138.0431 ms TTFT	232.4483 ms TTFT	361.4790 ms TTFT	524.71,129 ms TTFT	713.91,978 ms TTFT	Over capacity	Over capacity	Over capacity
64K	128.9601 ms TTFT	205.6803 ms TTFT	305.01,153 ms TTFT	416.11,692 ms TTFT	528.13,009 ms TTFT	Over capacity	Over capacity	Over capacity
128K	112.7826 ms TTFT	167.11,234 ms TTFT	232.61,755 ms TTFT	292.93,127 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	90.11,849 ms TTFT	123.32,589 ms TTFT	158.63,808 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	6,578.0	—	1,238 ms	—	—	1	—	—
Prefill	16K	1	Valid	6,556.0	—	2,458 ms	—	—	1	—	—
Prefill	32K	1	Valid	6,531.0	—	4,912 ms	—	—	1	—	—
Prefill	64K	1	Valid	5,621.0	—	11,390 ms	—	—	1	—	—
Prefill	128K	1	Valid	4,255.0	—	30,058 ms	—	—	1	—	—
Sustained decode	0	1	Valid	153.1	153.2	131 ms	6.5 ms	1.0	0	4,591	30.0 s
Sustained decode	0	2	Valid	265.8	133.5	230 ms	7.5 ms	2.0	0	7,974	30.0 s
Sustained decode	0	4	Valid	445.7	112.2	245 ms	8.9 ms	4.0	0	13,376	30.0 s
Sustained decode	0	8	Valid	721.0	90.7	352 ms	11.0 ms	8.0	0	21,648	30.0 s
Sustained decode	0	16	Valid	1,103.6	69.9	407 ms	14.3 ms	16.0	0	33,104	30.0 s
Sustained decode	0	32	Valid	1,603.9	51.2	639 ms	19.5 ms	32.0	0	48,128	30.0 s
Sustained decode	0	64	Valid	2,265.2	36.2	939 ms	27.6 ms	64.0	0	67,968	30.0 s
Sustained decode	0	128	Valid	2,936.7	23.2	1,250 ms	43.0 ms	128.0	0	88,064	30.0 s
Sustained decode	16K	1	Valid	147.4	147.5	291 ms	6.8 ms	1.0	0	4,421	30.0 s
Sustained decode	16K	2	Valid	247.1	123.6	417 ms	8.1 ms	2.0	0	7,412	30.0 s
Sustained decode	16K	4	Valid	403.6	100.6	618 ms	9.9 ms	4.0	0	12,108	30.0 s
Sustained decode	16K	8	Valid	603.2	74.7	922 ms	13.4 ms	8.0	0	18,112	30.0 s
Sustained decode	16K	16	Valid	854.8	52.3	1,709 ms	19.1 ms	16.0	0	25,648	30.0 s
Sustained decode	16K	32	Valid	1,157.5	34.8	3,082 ms	28.8 ms	32.0	0	34,720	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	138.0	138.4	431 ms	7.2 ms	1.0	0	4,140	30.0 s
Sustained decode	32K	2	Valid	232.4	116.4	483 ms	8.6 ms	2.0	0	6,970	30.0 s
Sustained decode	32K	4	Valid	361.4	90.2	790 ms	11.1 ms	4.0	0	10,852	30.0 s
Sustained decode	32K	8	Valid	524.7	65.0	1,129 ms	15.4 ms	8.0	0	15,752	30.0 s
Sustained decode	32K	16	Valid	713.9	43.6	1,978 ms	23.0 ms	16.0	0	21,408	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	128.9	129.4	601 ms	7.7 ms	1.0	0	3,865	30.0 s
Sustained decode	64K	2	Valid	205.6	103.0	803 ms	9.7 ms	2.0	0	6,168	30.0 s
Sustained decode	64K	4	Valid	305.0	75.9	1,153 ms	13.2 ms	4.0	0	9,148	30.0 s
Sustained decode	64K	8	Valid	416.1	51.1	1,692 ms	19.6 ms	8.0	0	12,488	30.0 s
Sustained decode	64K	16	Valid	528.1	31.5	3,009 ms	31.8 ms	16.0	0	15,840	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	112.7	113.2	826 ms	8.8 ms	1.0	0	3,379	30.0 s
Sustained decode	128K	2	Valid	167.1	84.0	1,234 ms	11.9 ms	2.0	0	5,016	30.0 s
Sustained decode	128K	4	Valid	232.6	58.2	1,755 ms	17.2 ms	4.0	0	6,976	30.0 s
Sustained decode	128K	8	Valid	292.9	36.5	3,127 ms	27.4 ms	8.0	0	8,784	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	90.1	90.7	1,849 ms	11.0 ms	1.0	0	2,702	30.0 s
Sustained decode	248K	2	Valid	123.3	61.9	2,589 ms	16.2 ms	2.0	0	3,698	30.0 s
Sustained decode	248K	4	Valid	158.6	39.3	3,808 ms	25.5 ms	4.0	0	4,756	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.