Speed test matrix

Fara1.5-4B-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, 10:19 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 prithivMLmods/Fara1.5-4B-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 context9,289877 ms TTFT

16K context10.3K1,572 ms TTFT

32K context9,3693,424 ms TTFT

64K context7,7148,300 ms TTFT

128K context5,60922,800 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	157.0104 ms TTFT	298.1190 ms TTFT	Looped	1,013.7242 ms TTFT	1,761.6308 ms TTFT	2,626.5418 ms TTFT	3,808.7668 ms TTFT	4,940.81,001 ms TTFT
16K	145.6217 ms TTFT	262.2261 ms TTFT	455.5386 ms TTFT	728.2633 ms TTFT	1,070.51,062 ms TTFT	1,351.71,777 ms TTFT	1,571.33,315 ms TTFT	Over capacity
32K	136.0301 ms TTFT	234.5362 ms TTFT	382.4603 ms TTFT	564.5679 ms TTFT	757.91,290 ms TTFT	893.52,293 ms TTFT	Over capacity	Over capacity
64K	120.8476 ms TTFT	192.8519 ms TTFT	288.9887 ms TTFT	388.51,330 ms TTFT	476.32,519 ms TTFT	Over capacity	Over capacity	Over capacity
128K	98.1764 ms TTFT	141.51,107 ms TTFT	193.21,556 ms TTFT	238.52,596 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	72.41,608 ms TTFT	94.22,060 ms TTFT	119.13,024 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

Looped: the model kept repeating the same text, so the benchmark discarded this measurement. 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	9,289.0	—	877 ms	—	—	1	—	—
Prefill	16K	1	Valid	10.3K	—	1,572 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,369.0	—	3,424 ms	—	—	1	—	—
Prefill	64K	1	Valid	7,714.0	—	8,300 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,609.0	—	22,800 ms	—	—	1	—	—
Sustained decode	0	1	Valid	157.0	157.5	104 ms	6.4 ms	1.0	0	4,711	30.0 s
Sustained decode	0	2	Valid	298.1	149.8	190 ms	6.7 ms	2.0	0	8,942	30.0 s
Sustained decode	0	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	8	Valid	1,013.7	128.3	242 ms	7.8 ms	8.0	0	30,440	30.0 s
Sustained decode	0	16	Valid	1,761.6	112.2	308 ms	8.9 ms	16.0	0	52,848	30.0 s
Sustained decode	0	32	Valid	2,626.5	83.8	418 ms	11.9 ms	32.0	0	78,816	30.0 s
Sustained decode	0	64	Valid	3,808.7	60.7	668 ms	16.5 ms	64.0	0	114,240	30.0 s
Sustained decode	0	128	Valid	4,940.8	38.9	1,001 ms	25.7 ms	128.0	0	148,352	30.0 s
Sustained decode	16K	1	Valid	145.6	146.0	217 ms	6.8 ms	1.0	0	4,367	30.0 s
Sustained decode	16K	2	Valid	262.2	131.5	261 ms	7.6 ms	2.0	0	7,866	30.0 s
Sustained decode	16K	4	Valid	455.5	114.0	386 ms	8.8 ms	4.0	0	13,664	30.0 s
Sustained decode	16K	8	Valid	728.2	91.0	633 ms	11.0 ms	8.0	0	21,848	30.0 s
Sustained decode	16K	16	Valid	1,070.5	66.5	1,062 ms	15.0 ms	16.0	0	32,112	30.0 s
Sustained decode	16K	32	Valid	1,351.7	41.5	1,777 ms	24.1 ms	32.0	0	40,544	30.0 s
Sustained decode	16K	64	Valid	1,571.3	23.3	3,315 ms	42.8 ms	64.0	0	47,168	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	136.0	136.5	301 ms	7.3 ms	1.0	0	4,080	30.0 s
Sustained decode	32K	2	Valid	234.5	117.5	362 ms	8.5 ms	2.0	0	7,034	30.0 s
Sustained decode	32K	4	Valid	382.4	95.7	603 ms	10.5 ms	4.0	0	11,472	30.0 s
Sustained decode	32K	8	Valid	564.5	70.2	679 ms	14.2 ms	8.0	0	16,936	30.0 s
Sustained decode	32K	16	Valid	757.9	46.7	1,290 ms	21.4 ms	16.0	0	22,736	30.0 s
Sustained decode	32K	32	Valid	893.5	26.9	2,293 ms	37.2 ms	32.0	0	26,784	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	120.8	121.3	476 ms	8.2 ms	1.0	0	3,625	30.0 s
Sustained decode	64K	2	Valid	192.8	96.7	519 ms	10.3 ms	2.0	0	5,784	30.0 s
Sustained decode	64K	4	Valid	288.9	72.3	887 ms	13.8 ms	4.0	0	8,664	30.0 s
Sustained decode	64K	8	Valid	388.5	48.4	1,330 ms	20.7 ms	8.0	0	11,656	30.0 s
Sustained decode	64K	16	Valid	476.3	29.5	2,519 ms	33.9 ms	16.0	0	14,288	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	98.1	98.5	764 ms	10.1 ms	1.0	0	2,944	30.0 s
Sustained decode	128K	2	Valid	141.5	71.1	1,107 ms	14.1 ms	2.0	0	4,244	30.0 s
Sustained decode	128K	4	Valid	193.2	48.5	1,556 ms	20.6 ms	4.0	0	5,796	30.0 s
Sustained decode	128K	8	Valid	238.5	29.7	2,596 ms	33.6 ms	8.0	0	7,152	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	1,608 ms	13.7 ms	1.0	0	2,173	30.0 s
Sustained decode	248K	2	Valid	94.2	47.6	2,060 ms	21.0 ms	2.0	0	2,826	30.0 s
Sustained decode	248K	4	Valid	119.1	30.0	3,024 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	—	—	—	—	—	—	—	—

Looped: the model kept repeating the same text, so the benchmark discarded this measurement. Over capacity: not run because this many requests at this context would exceed the server's KV cache.