Speed test matrix

Fara1.5-4B-AutoRound-W4A16-RTN on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · AutoRound W4A166 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, 01:41 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: AutoRound W4A16
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 LeaderboardModel1/Fara1.5-4B-AutoRound-W4A16-RTN --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 context9,791832 ms TTFT

16K context9,5261,692 ms TTFT

32K context8,6993,688 ms TTFT

64K context7,1988,894 ms TTFT

128K context5,32024,041 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	191.9106 ms TTFT	363.1185 ms TTFT	676.6196 ms TTFT	1,199.1262 ms TTFT	1,933.1300 ms TTFT	2,928.0447 ms TTFT	Looped	4,682.71,003 ms TTFT
16K	175.2203 ms TTFT	314.5276 ms TTFT	534.6401 ms TTFT	828.0608 ms TTFT	1,133.4996 ms TTFT	1,433.71,855 ms TTFT	1,571.73,404 ms TTFT	Over capacity
32K	161.9281 ms TTFT	276.4382 ms TTFT	437.6515 ms TTFT	625.4800 ms TTFT	791.41,259 ms TTFT	931.12,241 ms TTFT	Over capacity	Over capacity
64K	140.8481 ms TTFT	220.7643 ms TTFT	320.6860 ms TTFT	416.61,385 ms TTFT	489.52,381 ms TTFT	Over capacity	Over capacity	Over capacity
128K	111.2786 ms TTFT	156.21,111 ms TTFT	207.31,500 ms TTFT	249.32,330 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	79.21,557 ms TTFT	100.62,034 ms TTFT	124.53,016 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	9,791.0	—	832 ms	—	—	1	—	—
Prefill	16K	1	Valid	9,526.0	—	1,692 ms	—	—	1	—	—
Prefill	32K	1	Valid	8,699.0	—	3,688 ms	—	—	1	—	—
Prefill	64K	1	Valid	7,198.0	—	8,894 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,320.0	—	24,041 ms	—	—	1	—	—
Sustained decode	0	1	Valid	191.9	192.6	106 ms	5.2 ms	1.0	0	5,756	30.0 s
Sustained decode	0	2	Valid	363.1	182.6	185 ms	5.5 ms	2.0	0	10,892	30.0 s
Sustained decode	0	4	Valid	676.6	171.0	196 ms	5.8 ms	4.0	0	20,296	30.0 s
Sustained decode	0	8	Valid	1,199.1	152.3	262 ms	6.6 ms	8.0	0	35,976	30.0 s
Sustained decode	0	16	Valid	1,933.1	123.6	300 ms	8.1 ms	16.0	0	58,000	30.0 s
Sustained decode	0	32	Valid	2,928.0	93.9	447 ms	10.7 ms	32.0	0	87,840	30.0 s
Sustained decode	0	64	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	128	Valid	4,682.7	36.7	1,003 ms	27.2 ms	128.0	0	140,544	30.0 s
Sustained decode	16K	1	Valid	175.2	175.7	203 ms	5.7 ms	1.0	0	5,256	30.0 s
Sustained decode	16K	2	Valid	314.5	157.7	276 ms	6.3 ms	2.0	0	9,434	30.0 s
Sustained decode	16K	4	Valid	534.6	134.0	401 ms	7.5 ms	4.0	0	16,038	30.0 s
Sustained decode	16K	8	Valid	828.0	103.4	608 ms	9.7 ms	8.0	0	24,860	30.0 s
Sustained decode	16K	16	Valid	1,133.4	70.3	996 ms	14.2 ms	16.0	0	34,000	30.0 s
Sustained decode	16K	32	Valid	1,433.7	43.9	1,855 ms	22.8 ms	32.0	0	43,008	30.0 s
Sustained decode	16K	64	Valid	1,571.7	23.3	3,404 ms	43.0 ms	64.0	0	47,168	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	161.9	162.4	281 ms	6.2 ms	1.0	0	4,857	30.0 s
Sustained decode	32K	2	Valid	276.4	138.7	382 ms	7.2 ms	2.0	0	8,292	30.0 s
Sustained decode	32K	4	Valid	437.6	109.5	515 ms	9.1 ms	4.0	0	13,128	30.0 s
Sustained decode	32K	8	Valid	625.4	77.9	800 ms	12.8 ms	8.0	0	18,776	30.0 s
Sustained decode	32K	16	Valid	791.4	48.8	1,259 ms	20.5 ms	16.0	0	23,744	30.0 s
Sustained decode	32K	32	Valid	931.1	27.9	2,241 ms	35.8 ms	32.0	0	27,968	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	140.8	141.4	481 ms	7.1 ms	1.0	0	4,224	30.0 s
Sustained decode	64K	2	Valid	220.7	110.8	643 ms	9.0 ms	2.0	0	6,620	30.0 s
Sustained decode	64K	4	Valid	320.6	80.2	860 ms	12.5 ms	4.0	0	9,619	30.0 s
Sustained decode	64K	8	Valid	416.6	51.9	1,385 ms	19.3 ms	8.0	0	12,496	30.0 s
Sustained decode	64K	16	Valid	489.5	30.3	2,381 ms	33.0 ms	16.0	0	14,672	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	111.2	111.8	786 ms	8.9 ms	1.0	0	3,337	30.0 s
Sustained decode	128K	2	Valid	156.2	78.5	1,111 ms	12.7 ms	2.0	0	4,686	30.0 s
Sustained decode	128K	4	Valid	207.3	52.0	1,500 ms	19.2 ms	4.0	0	6,216	30.0 s
Sustained decode	128K	8	Valid	249.3	31.1	2,330 ms	32.1 ms	8.0	0	7,472	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	79.2	79.8	1,557 ms	12.5 ms	1.0	0	2,376	30.0 s
Sustained decode	248K	2	Valid	100.6	50.8	2,034 ms	19.7 ms	2.0	0	3,018	30.0 s
Sustained decode	248K	4	Valid	124.5	31.4	3,016 ms	31.9 ms	4.0	0	3,732	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.