Speed test matrix

Qwen3-235B-A22B-Instruct-2507-REAP-nvfp4 on CMP 170HX 64GB ×2

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 31, 2026, 20:31 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: 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: none. KV cache: int4_per_token_head. 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 Banana-Bae/Qwen3-235B-A22B-Instruct-2507-REAP-nvfp4 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --pipeline-parallel-size 2 --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context1,0487,831 ms TTFT

16K context764.021,259 ms TTFT

32K context480.067,374 ms TTFT

64K context272.0237,323 ms TTFT

128K context347.0371,491 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	39.846,761 ms TTFT	65.443,728 ms TTFT	116.4323 ms TTFT	154.0344 ms TTFT	240.1501 ms TTFT	366.1654 ms TTFT	Over capacity	Over capacity
16K	23.9326 ms TTFT	45.4382 ms TTFT	68.5571 ms TTFT	100.0538 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	17.0Requests queued533 ms TTFT	32.5670 ms TTFT	49.7866 ms TTFT	71.71,264 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	10.7Requests queued936 ms TTFT	20.71,272 ms TTFT	32.61,603 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	Over capacity	11.1Requests queued2,421 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	Over capacity	Over capacity	Over capacity	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	1,048.0	—	7,831 ms	—	—	1	—	—
Prefill	16K	1	Valid	764.0	—	21,259 ms	—	—	1	—	—
Prefill	32K	1	Valid	480.0	—	67,374 ms	—	—	1	—	—
Prefill	64K	1	Valid	272.0	—	237,323 ms	—	—	1	—	—
Prefill	128K	1	Valid	347.0	—	371,491 ms	—	—	1	—	—
Sustained decode	0	1	Valid	39.8	40.0	46,761 ms	25.0 ms	1.0	0	1,195	30.0 s
Sustained decode	0	2	Valid	65.4	32.9	43,728 ms	30.4 ms	2.0	0	1,960	30.0 s
Sustained decode	0	4	Valid	116.4	29.3	323 ms	34.1 ms	4.0	0	3,490	30.0 s
Sustained decode	0	8	Valid	154.0	19.9	344 ms	50.1 ms	8.0	0	4,616	30.0 s
Sustained decode	0	16	Valid	240.1	15.5	501 ms	64.4 ms	16.0	0	7,200	30.0 s
Sustained decode	0	32	Valid	366.1	11.8	654 ms	84.6 ms	32.0	0	10,976	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	23.9	24.0	326 ms	41.7 ms	1.0	0	718	30.0 s
Sustained decode	16K	2	Valid	45.4	22.7	382 ms	44.0 ms	2.0	0	1,362	30.0 s
Sustained decode	16K	4	Valid	68.5	17.2	571 ms	58.0 ms	4.0	0	2,053	30.0 s
Sustained decode	16K	8	Valid	100.0	12.7	538 ms	79.0 ms	8.0	0	3,000	30.0 s
Sustained decode	16K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Requests queued	17.0	17.0	533 ms	58.7 ms	1.0	0	510	30.0 s
Sustained decode	32K	2	Valid	32.5	16.2	670 ms	61.6 ms	2.0	0	974	30.0 s
Sustained decode	32K	4	Valid	49.7	12.4	866 ms	80.3 ms	4.0	0	1,489	30.0 s
Sustained decode	32K	8	Valid	71.7	9.0	1,264 ms	111.2 ms	8.0	0	2,147	29.9 s
Sustained decode	32K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Requests queued	10.7	10.8	936 ms	92.7 ms	1.0	0	320	30.0 s
Sustained decode	64K	2	Valid	20.7	10.3	1,272 ms	97.1 ms	2.0	0	620	30.0 s
Sustained decode	64K	4	Valid	32.6	8.1	1,603 ms	123.6 ms	4.0	0	976	29.9 s
Sustained decode	64K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Over capacity	—	—	—	—	0.5	0	0	30.0 s
Sustained decode	128K	2	Requests queued	11.1	5.9	2,421 ms	169.2 ms	1.9	0	333	30.0 s
Sustained decode	128K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	8	Over capacity	—	—	—	—	—	—	—	—
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	Over capacity	—	—	—	—	0.6	0	0	30.0 s
Sustained decode	248K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	4	Over capacity	—	—	—	—	—	—	—	—
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.