Speed test matrix

Qwen3.6-35B-A3B-NVFP4 on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · NVFP46 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 03:01 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: NVFP4
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 nvidia/Qwen3.6-35B-A3B-NVFP4 --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 context5,0341,618 ms TTFT

16K context6,6392,428 ms TTFT

32K context6,4804,951 ms TTFT

64K context5,56611,502 ms TTFT

128K context4,12930,973 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	215.7164 ms TTFT	395.8279 ms TTFT	657.6311 ms TTFT	960.0443 ms TTFT	1,370.7467 ms TTFT	1,770.4627 ms TTFT	2,304.41,084 ms TTFT	2,916.21,427 ms TTFT
16K	130.1279 ms TTFT	245.8480 ms TTFT	429.2912 ms TTFT	527.81,122 ms TTFT	630.61,912 ms TTFT	740.53,441 ms TTFT	799.66,952 ms TTFT	Over capacity
32K	90.0538 ms TTFT	170.7832 ms TTFT	306.51,303 ms TTFT	357.82,134 ms TTFT	407.83,899 ms TTFT	461.57,822 ms TTFT	Over capacity	Over capacity
64K	54.5750 ms TTFT	109.41,069 ms TTFT	195.01,760 ms TTFT	215.82,710 ms TTFT	233.44,621 ms TTFT	Over capacity	Over capacity	Over capacity
128K	30.61,242 ms TTFT	60.61,639 ms TTFT	113.92,430 ms TTFT	121.63,669 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	17.32,099 ms TTFT	34.72,873 ms TTFT	63.34,119 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	5,034.0	—	1,618 ms	—	—	1	—	—
Prefill	16K	1	Valid	6,639.0	—	2,428 ms	—	—	1	—	—
Prefill	32K	1	Valid	6,480.0	—	4,951 ms	—	—	1	—	—
Prefill	64K	1	Valid	5,566.0	—	11,502 ms	—	—	1	—	—
Prefill	128K	1	Valid	4,129.0	—	30,973 ms	—	—	1	—	—
Sustained decode	0	1	Valid	215.7	222.6	164 ms	4.5 ms	1.0	0	6,472	30.0 s
Sustained decode	0	2	Valid	395.8	202.6	279 ms	4.9 ms	2.0	0	11,870	30.0 s
Sustained decode	0	4	Valid	657.6	167.7	311 ms	6.0 ms	4.0	0	19,728	30.0 s
Sustained decode	0	8	Valid	960.0	123.6	443 ms	8.1 ms	8.0	0	28,799	30.0 s
Sustained decode	0	16	Valid	1,370.7	88.6	467 ms	11.3 ms	16.0	0	41,142	30.0 s
Sustained decode	0	32	Valid	1,770.4	57.1	627 ms	17.5 ms	32.0	0	53,138	30.0 s
Sustained decode	0	64	Valid	2,304.4	37.4	1,084 ms	26.7 ms	64.0	0	69,077	30.0 s
Sustained decode	0	128	Valid	2,916.2	23.6	1,427 ms	42.4 ms	128.0	0	87,351	30.0 s
Sustained decode	16K	1	Valid	130.1	131.6	279 ms	7.6 ms	1.0	0	3,902	30.0 s
Sustained decode	16K	2	Valid	245.8	123.8	480 ms	8.1 ms	2.0	0	7,373	30.0 s
Sustained decode	16K	4	Valid	429.2	107.3	912 ms	9.3 ms	4.0	0	12,877	30.0 s
Sustained decode	16K	8	Valid	527.8	65.8	1,122 ms	15.2 ms	8.0	0	15,825	30.0 s
Sustained decode	16K	16	Valid	630.6	39.6	1,912 ms	25.3 ms	16.0	0	18,910	30.0 s
Sustained decode	16K	32	Valid	740.5	22.8	3,441 ms	43.9 ms	32.0	0	22,203	30.0 s
Sustained decode	16K	64	Valid	799.6	12.2	6,952 ms	82.1 ms	64.0	0	23,996	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	90.0	91.5	538 ms	10.9 ms	1.0	0	2,699	30.0 s
Sustained decode	32K	2	Valid	170.7	86.4	832 ms	11.6 ms	2.0	0	5,119	30.0 s
Sustained decode	32K	4	Valid	306.5	76.3	1,303 ms	13.1 ms	4.0	0	9,196	30.0 s
Sustained decode	32K	8	Valid	357.8	44.2	2,134 ms	22.6 ms	8.0	0	10,730	30.0 s
Sustained decode	32K	16	Valid	407.8	24.3	3,899 ms	41.1 ms	16.0	0	12,202	29.9 s
Sustained decode	32K	32	Valid	461.5	13.6	7,822 ms	73.3 ms	32.0	0	13,795	29.9 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	54.5	55.1	750 ms	18.2 ms	1.0	0	1,633	30.0 s
Sustained decode	64K	2	Valid	109.4	55.1	1,069 ms	18.2 ms	2.0	0	3,280	30.0 s
Sustained decode	64K	4	Valid	195.0	48.9	1,760 ms	20.4 ms	4.0	0	5,844	30.0 s
Sustained decode	64K	8	Valid	215.8	26.6	2,710 ms	37.7 ms	8.0	0	6,463	30.0 s
Sustained decode	64K	16	Valid	233.4	13.7	4,621 ms	72.7 ms	16.0	0	6,994	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	30.6	31.6	1,242 ms	31.7 ms	1.0	0	917	30.0 s
Sustained decode	128K	2	Valid	60.6	30.8	1,639 ms	32.5 ms	2.0	0	1,816	29.9 s
Sustained decode	128K	4	Valid	113.9	28.6	2,430 ms	34.9 ms	4.0	0	3,412	30.0 s
Sustained decode	128K	8	Valid	121.6	15.3	3,669 ms	65.2 ms	8.0	0	3,642	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	17.3	17.8	2,099 ms	56.1 ms	1.0	0	520	30.0 s
Sustained decode	248K	2	Valid	34.7	17.7	2,873 ms	56.6 ms	2.0	0	1,041	30.0 s
Sustained decode	248K	4	Valid	63.3	16.1	4,119 ms	62.0 ms	4.0	0	1,892	29.9 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.