Speed test matrix

Agents-A1-4B-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, 09:52 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 InternScience/Agents-A1-4B-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 context10.8K777 ms TTFT

16K context10.3K1,594 ms TTFT

32K context9,2153,508 ms TTFT

64K context7,6718,378 ms TTFT

128K context5,60722,851 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	156.6135 ms TTFT	297.6169 ms TTFT	560.2241 ms TTFT	1,005.6391 ms TTFT	1,745.2485 ms TTFT	2,594.4728 ms TTFT	3,748.21,230 ms TTFT	4,812.82,156 ms TTFT
16K	145.5226 ms TTFT	262.3292 ms TTFT	454.9423 ms TTFT	724.5587 ms TTFT	1,061.41,042 ms TTFT	1,339.21,911 ms TTFT	1,557.03,888 ms TTFT	Over capacity
32K	136.4312 ms TTFT	234.3359 ms TTFT	382.2563 ms TTFT	562.7824 ms TTFT	755.71,550 ms TTFT	891.22,621 ms TTFT	Over capacity	Over capacity
64K	121.1528 ms TTFT	192.7599 ms TTFT	288.6946 ms TTFT	388.11,471 ms TTFT	475.22,566 ms TTFT	Over capacity	Over capacity	Over capacity
128K	98.5996 ms TTFT	141.41,236 ms TTFT	193.11,714 ms TTFT	238.22,697 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	72.41,785 ms TTFT	94.22,217 ms TTFT	119.13,274 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	10.8K	—	777 ms	—	—	1	—	—
Prefill	16K	1	Valid	10.3K	—	1,594 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,215.0	—	3,508 ms	—	—	1	—	—
Prefill	64K	1	Valid	7,671.0	—	8,378 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,607.0	—	22,851 ms	—	—	1	—	—
Sustained decode	0	1	Valid	156.6	157.3	135 ms	6.4 ms	1.0	0	4,699	30.0 s
Sustained decode	0	2	Valid	297.6	149.5	169 ms	6.7 ms	2.0	0	8,926	30.0 s
Sustained decode	0	4	Valid	560.2	141.1	241 ms	7.1 ms	4.0	0	16,818	30.0 s
Sustained decode	0	8	Valid	1,005.6	127.2	391 ms	7.9 ms	8.0	0	30,169	30.0 s
Sustained decode	0	16	Valid	1,745.2	110.8	485 ms	9.0 ms	16.0	0	52,356	30.0 s
Sustained decode	0	32	Valid	2,594.4	81.9	728 ms	12.2 ms	32.0	0	77,824	30.0 s
Sustained decode	0	64	Valid	3,748.2	58.4	1,230 ms	17.1 ms	64.0	0	112,448	30.0 s
Sustained decode	0	128	Valid	4,812.8	36.6	2,156 ms	27.3 ms	128.0	0	144,411	30.0 s
Sustained decode	16K	1	Valid	145.5	145.8	226 ms	6.9 ms	1.0	0	4,367	30.0 s
Sustained decode	16K	2	Valid	262.3	131.5	292 ms	7.6 ms	2.0	0	7,870	30.0 s
Sustained decode	16K	4	Valid	454.9	114.0	423 ms	8.8 ms	4.0	0	13,648	30.0 s
Sustained decode	16K	8	Valid	724.5	90.4	587 ms	11.1 ms	8.0	0	21,752	30.0 s
Sustained decode	16K	16	Valid	1,061.4	65.7	1,042 ms	15.2 ms	16.0	0	31,840	30.0 s
Sustained decode	16K	32	Valid	1,339.2	40.7	1,911 ms	24.6 ms	32.0	0	40,160	30.0 s
Sustained decode	16K	64	Valid	1,557.0	22.6	3,888 ms	44.2 ms	64.0	0	46,720	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	136.4	136.9	312 ms	7.3 ms	1.0	0	4,091	30.0 s
Sustained decode	32K	2	Valid	234.3	117.5	359 ms	8.5 ms	2.0	0	7,026	30.0 s
Sustained decode	32K	4	Valid	382.2	95.7	563 ms	10.4 ms	4.0	0	11,464	30.0 s
Sustained decode	32K	8	Valid	562.7	70.2	824 ms	14.2 ms	8.0	0	16,880	30.0 s
Sustained decode	32K	16	Valid	755.7	47.0	1,550 ms	21.3 ms	16.0	0	22,672	30.0 s
Sustained decode	32K	32	Valid	891.2	27.3	2,621 ms	36.6 ms	32.0	0	26,720	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	121.1	121.5	528 ms	8.2 ms	1.0	0	3,633	30.0 s
Sustained decode	64K	2	Valid	192.7	96.7	599 ms	10.3 ms	2.0	0	5,782	30.0 s
Sustained decode	64K	4	Valid	288.6	72.2	946 ms	13.9 ms	4.0	0	8,656	30.0 s
Sustained decode	64K	8	Valid	388.1	48.2	1,471 ms	20.8 ms	8.0	0	11,640	30.0 s
Sustained decode	64K	16	Valid	475.2	29.2	2,566 ms	34.3 ms	16.0	0	14,256	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	98.5	99.0	996 ms	10.1 ms	1.0	0	2,954	30.0 s
Sustained decode	128K	2	Valid	141.4	71.0	1,236 ms	14.1 ms	2.0	0	4,242	30.0 s
Sustained decode	128K	4	Valid	193.1	48.2	1,714 ms	20.7 ms	4.0	0	5,792	30.0 s
Sustained decode	128K	8	Valid	238.2	29.6	2,697 ms	33.8 ms	8.0	0	7,144	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	72.4	72.9	1,785 ms	13.7 ms	1.0	0	2,172	30.0 s
Sustained decode	248K	2	Valid	94.2	47.5	2,217 ms	21.0 ms	2.0	0	2,827	30.0 s
Sustained decode	248K	4	Valid	119.1	29.9	3,274 ms	33.4 ms	4.0	0	3,572	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.