Speed test matrix

Llama-3.2-3B-Instruct on RTX 3090 24GB

vllm · 0.32.0 · BF165 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 05:57 UTC

Hardware

Type: Discrete GPU
GPU: RTX 3090
VRAM: 24 GB
CPU: Intel Core i7-9700K (8C)
RAM: 32 GB
OS: Ubuntu 24.04.5

Engine

Engine: vllm
Version: 0.32.0
Quantization: BF16
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: vllm/vllm-openai:v0.32.0-x86_64

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

# Model files are alpindale/Llama-3.2-3B-Instruct, a byte-identical mirror of meta-llama/Llama-3.2-3B-Instruct (all weight, config and tokenizer files match the official repo sizes; only README differs).
# Untuned out-of-the-box baseline. Spec decode: none
# 1x RTX 3090 24GB (Zagitta/gpu-unlock), 420 W limit, PCIe Gen3 x16 | Intel Core i7-9700K 8C | 32 GB DDR4 | Ubuntu 24.04.5, NVIDIA driver 580.178.04
vllm serve alpindale/Llama-3.2-3B-Instruct --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 context14.7K560 ms TTFT

16K context12.2K1,335 ms TTFT

32K context9,1383,554 ms TTFT

64K context6,00010,797 ms TTFT

120K context3,73132,518 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	110.118 ms TTFT	210.221 ms TTFT	397.634 ms TTFT	716.134 ms TTFT	1,232.049 ms TTFT	Over capacity	Over capacity	Over capacity
16K	89.560 ms TTFT	151.180 ms TTFT	268.0119 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
32K	75.588 ms TTFT	117.2121 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	57.2217 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	40.2431 ms TTFT	Over capacity	Over capacity	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	14.7K	—	560 ms	—	—	1	—	—
Prefill	16K	1	Valid	12.2K	—	1,335 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,138.0	—	3,554 ms	—	—	1	—	—
Prefill	64K	1	Valid	6,000.0	—	10,797 ms	—	—	1	—	—
Prefill	120K	1	Valid	3,731.0	—	32,518 ms	—	—	1	—	—
Sustained decode	0	1	Valid	110.1	110.7	18 ms	9.0 ms	1.0	0	3,303	30.0 s
Sustained decode	0	2	Valid	210.2	106.2	21 ms	9.4 ms	2.0	0	6,306	30.0 s
Sustained decode	0	4	Valid	397.6	101.1	34 ms	9.9 ms	4.0	0	11,929	30.0 s
Sustained decode	0	8	Valid	716.1	92.1	34 ms	10.9 ms	8.0	0	21,480	30.0 s
Sustained decode	0	16	Valid	1,232.0	80.7	49 ms	12.4 ms	16.0	0	36,960	30.0 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	89.5	89.6	60 ms	11.2 ms	1.0	0	2,685	30.0 s
Sustained decode	16K	2	Valid	151.1	75.7	80 ms	13.2 ms	2.0	0	4,530	30.0 s
Sustained decode	16K	4	Valid	268.0	67.4	119 ms	14.8 ms	4.0	0	8,039	30.0 s
Sustained decode	16K	8	Over capacity	—	—	—	—	—	—	—	—
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	Valid	75.5	75.6	88 ms	13.2 ms	1.0	0	2,264	30.0 s
Sustained decode	32K	2	Valid	117.2	58.7	121 ms	17.0 ms	2.0	0	3,516	30.0 s
Sustained decode	32K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	8	Over capacity	—	—	—	—	—	—	—	—
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	57.2	57.3	217 ms	17.4 ms	1.0	0	1,717	30.0 s
Sustained decode	64K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	4	Over capacity	—	—	—	—	—	—	—	—
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	40.2	40.3	431 ms	24.8 ms	1.0	0	1,206	30.0 s
Sustained decode	120K	2	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.