Speed test matrix

Llama-3.2-3B-Instruct-FP8-dynamic on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP8 dynamic5 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 10:36 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 dynamic
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 RedHatAI/Llama-3.2-3B-Instruct-FP8-dynamic --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context13.6K601 ms TTFT

16K context11.6K1,395 ms TTFT

32K context9,0183,578 ms TTFT

64K context6,14310,480 ms TTFT

120K context3,90830,847 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	203.456 ms TTFT	Looped	654.098 ms TTFT	Looped	Looped	2,507.0202 ms TTFT	Over capacity	Over capacity
16K	156.9137 ms TTFT	250.5203 ms TTFT	374.2316 ms TTFT	500.7428 ms TTFT	610.5716 ms TTFT	Over capacity	Over capacity	Over capacity
32K	126.2230 ms TTFT	183.5340 ms TTFT	246.4500 ms TTFT	302.8810 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	90.6451 ms TTFT	116.3608 ms TTFT	144.5848 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	60.4868 ms TTFT	71.01,060 ms TTFT	Over capacity	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	13.6K	—	601 ms	—	—	1	—	—
Prefill	16K	1	Valid	11.6K	—	1,395 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,018.0	—	3,578 ms	—	—	1	—	—
Prefill	64K	1	Valid	6,143.0	—	10,480 ms	—	—	1	—	—
Prefill	120K	1	Valid	3,908.0	—	30,847 ms	—	—	1	—	—
Sustained decode	0	1	Valid	203.4	206.4	56 ms	4.8 ms	1.0	0	6,102	30.0 s
Sustained decode	0	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	4	Valid	654.0	170.0	98 ms	5.9 ms	4.0	0	19,620	30.0 s
Sustained decode	0	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	32	Valid	2,507.0	86.5	202 ms	11.6 ms	32.0	0	75,232	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	156.9	157.8	137 ms	6.3 ms	1.0	0	4,706	30.0 s
Sustained decode	16K	2	Valid	250.5	126.4	203 ms	7.9 ms	2.0	0	7,514	30.0 s
Sustained decode	16K	4	Valid	374.2	94.4	316 ms	10.6 ms	4.0	0	11,225	30.0 s
Sustained decode	16K	8	Valid	500.7	63.1	428 ms	15.9 ms	8.0	0	15,016	30.0 s
Sustained decode	16K	16	Valid	610.5	38.2	716 ms	26.2 ms	16.0	0	18,304	30.0 s
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	126.2	126.9	230 ms	7.9 ms	1.0	0	3,786	30.0 s
Sustained decode	32K	2	Valid	183.5	92.2	340 ms	10.8 ms	2.0	0	5,504	30.0 s
Sustained decode	32K	4	Valid	246.4	62.0	500 ms	16.1 ms	4.0	0	7,392	30.0 s
Sustained decode	32K	8	Valid	302.8	38.0	810 ms	26.3 ms	8.0	0	9,080	30.0 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	Valid	90.6	91.0	451 ms	11.0 ms	1.0	0	2,717	30.0 s
Sustained decode	64K	2	Valid	116.3	58.4	608 ms	17.1 ms	2.0	0	3,488	30.0 s
Sustained decode	64K	4	Valid	144.5	36.3	848 ms	27.6 ms	4.0	0	4,336	30.0 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	120K	1	Valid	60.4	60.7	868 ms	16.5 ms	1.0	0	1,811	30.0 s
Sustained decode	120K	2	Valid	71.0	35.7	1,060 ms	28.0 ms	2.0	0	2,130	30.0 s
Sustained decode	120K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	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.