Speed test matrix

Qwen3-Next-80B-A3B-Instruct-FP8 on CMP 170HX 64GB ×2

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 31, 2026, 04:24 UTC

Hardware

Type: Discrete GPU
GPU: CMP 170HX
GPU count: 2
VRAM: 64 GB (2 × 32 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. Layout: PP2 (stock vLLM flag)
# 2x 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-Next-80B-A3B-Instruct-FP8 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --pipeline-parallel-size 2

Prefill

prompt tok/s by context length

8K context9,334879 ms TTFT

16K context10.8K1,504 ms TTFT

32K context10.5K3,087 ms TTFT

64K context9,1177,078 ms TTFT

128K context7,02318,356 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	123.2137 ms TTFT	212.8138 ms TTFT	379.4217 ms TTFT	544.0376 ms TTFT	822.4407 ms TTFT	1,223.2556 ms TTFT	1,968.9790 ms TTFT	2,821.51,031 ms TTFT
16K	117.7338 ms TTFT	208.2451 ms TTFT	358.8597 ms TTFT	572.0943 ms TTFT	851.21,586 ms TTFT	1,205.02,854 ms TTFT	1,653.15,350 ms TTFT	Over capacity
32K	111.9335 ms TTFT	187.8381 ms TTFT	306.7506 ms TTFT	524.9774 ms TTFT	743.71,276 ms TTFT	1,030.62,352 ms TTFT	Over capacity	Over capacity
64K	105.0598 ms TTFT	171.2685 ms TTFT	263.11,049 ms TTFT	455.01,616 ms TTFT	623.32,610 ms TTFT	Over capacity	Over capacity	Over capacity
128K	93.8968 ms TTFT	168.81,266 ms TTFT	205.71,777 ms TTFT	266.32,904 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	77.91,897 ms TTFT	138.02,344 ms TTFT	144.13,363 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	9,334.0	—	879 ms	—	—	1	—	—
Prefill	16K	1	Valid	10.8K	—	1,504 ms	—	—	1	—	—
Prefill	32K	1	Valid	10.5K	—	3,087 ms	—	—	1	—	—
Prefill	64K	1	Valid	9,117.0	—	7,078 ms	—	—	1	—	—
Prefill	128K	1	Valid	7,023.0	—	18,356 ms	—	—	1	—	—
Sustained decode	0	1	Valid	123.2	123.2	137 ms	8.1 ms	1.0	0	3,694	30.0 s
Sustained decode	0	2	Valid	212.8	106.8	138 ms	9.4 ms	2.0	0	6,383	30.0 s
Sustained decode	0	4	Valid	379.4	95.1	217 ms	10.5 ms	4.0	0	11,382	30.0 s
Sustained decode	0	8	Valid	544.0	69.2	376 ms	14.5 ms	8.0	0	16,320	30.0 s
Sustained decode	0	16	Valid	822.4	52.8	407 ms	18.9 ms	16.0	0	24,673	30.0 s
Sustained decode	0	32	Valid	1,223.2	39.5	556 ms	25.3 ms	32.0	0	36,704	30.0 s
Sustained decode	0	64	Valid	1,968.9	32.0	790 ms	31.2 ms	64.0	0	59,072	30.0 s
Sustained decode	0	128	Valid	2,821.5	22.8	1,031 ms	43.9 ms	128.0	0	84,681	30.0 s
Sustained decode	16K	1	Valid	117.7	117.8	338 ms	8.5 ms	1.0	0	3,529	30.0 s
Sustained decode	16K	2	Valid	208.2	103.6	451 ms	9.7 ms	2.0	0	6,244	30.0 s
Sustained decode	16K	4	Valid	358.8	89.3	597 ms	11.2 ms	4.0	0	10,762	30.0 s
Sustained decode	16K	8	Valid	572.0	70.8	943 ms	14.1 ms	8.0	0	17,156	30.0 s
Sustained decode	16K	16	Valid	851.2	52.1	1,586 ms	19.2 ms	16.0	0	25,530	30.0 s
Sustained decode	16K	32	Valid	1,205.0	35.9	2,854 ms	27.9 ms	32.0	0	36,144	30.0 s
Sustained decode	16K	64	Valid	1,653.1	23.5	5,350 ms	42.5 ms	64.0	0	49,600	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	111.9	112.2	335 ms	8.9 ms	1.0	0	3,355	30.0 s
Sustained decode	32K	2	Valid	187.8	94.1	381 ms	10.6 ms	2.0	0	5,634	30.0 s
Sustained decode	32K	4	Valid	306.7	76.6	506 ms	13.0 ms	4.0	0	9,201	30.0 s
Sustained decode	32K	8	Valid	524.9	65.4	774 ms	15.3 ms	8.0	0	15,744	30.0 s
Sustained decode	32K	16	Valid	743.7	46.0	1,276 ms	21.8 ms	16.0	0	22,311	30.0 s
Sustained decode	32K	32	Valid	1,030.6	31.2	2,352 ms	32.1 ms	32.0	0	30,928	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	105.0	105.2	598 ms	9.5 ms	1.0	0	3,150	30.0 s
Sustained decode	64K	2	Valid	171.2	85.6	685 ms	11.7 ms	2.0	0	5,134	30.0 s
Sustained decode	64K	4	Valid	263.1	65.5	1,049 ms	15.3 ms	4.0	0	7,892	30.0 s
Sustained decode	64K	8	Valid	455.0	56.0	1,616 ms	17.9 ms	8.0	0	13,647	30.0 s
Sustained decode	64K	16	Valid	623.3	37.5	2,610 ms	26.7 ms	16.0	0	18,696	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	93.8	94.1	968 ms	10.6 ms	1.0	0	2,814	30.0 s
Sustained decode	128K	2	Valid	168.8	84.4	1,266 ms	11.8 ms	2.0	0	5,062	30.0 s
Sustained decode	128K	4	Valid	205.7	51.1	1,777 ms	19.6 ms	4.0	0	6,169	30.0 s
Sustained decode	128K	8	Valid	266.3	32.7	2,904 ms	30.6 ms	8.0	0	7,984	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	77.9	78.4	1,897 ms	12.7 ms	1.0	0	2,336	30.0 s
Sustained decode	248K	2	Valid	138.0	69.2	2,344 ms	14.5 ms	2.0	0	4,140	30.0 s
Sustained decode	248K	4	Valid	144.1	36.2	3,363 ms	27.6 ms	4.0	0	4,321	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.