Speed test matrix

Ministral-3-3B-Instruct-2512-FP8 on RTX 3090 24GB

vllm · 0.32.0 · FP86 contexts (0–160K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 06:44 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: FP8
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

# 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 unsloth/Ministral-3-3B-Instruct-2512-FP8 --max-model-len 172032 --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 context12.3K680 ms TTFT

16K context10.1K1,643 ms TTFT

32K context7,4244,453 ms TTFT

64K context4,88613,499 ms TTFT

128K context2,87145,900 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	159.619 ms TTFT	299.221 ms TTFT	545.422 ms TTFT	932.230 ms TTFT	1,524.067 ms TTFT	Over capacity	Over capacity	Over capacity
16K	122.348 ms TTFT	198.357 ms TTFT	343.286 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
32K	98.949 ms TTFT	146.269 ms TTFT	239.9108 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	71.1120 ms TTFT	95.6182 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	45.4248 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
160K	38.4297 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	12.3K	—	680 ms	—	—	1	—	—
Prefill	16K	1	Valid	10.1K	—	1,643 ms	—	—	1	—	—
Prefill	32K	1	Valid	7,424.0	—	4,453 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,886.0	—	13,499 ms	—	—	1	—	—
Prefill	128K	1	Valid	2,871.0	—	45,900 ms	—	—	1	—	—
Sustained decode	0	1	Valid	159.6	161.5	19 ms	6.2 ms	1.0	0	4,788	30.0 s
Sustained decode	0	2	Valid	299.2	152.5	21 ms	6.6 ms	2.0	0	8,975	30.0 s
Sustained decode	0	4	Valid	545.4	140.7	22 ms	7.1 ms	4.0	0	16,363	30.0 s
Sustained decode	0	8	Valid	932.2	122.7	30 ms	8.2 ms	8.0	0	27,967	30.0 s
Sustained decode	0	16	Valid	1,524.0	102.9	67 ms	9.7 ms	16.0	0	45,712	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	122.3	122.7	48 ms	8.1 ms	1.0	0	3,669	30.0 s
Sustained decode	16K	2	Valid	198.3	99.6	57 ms	10.0 ms	2.0	0	5,948	30.0 s
Sustained decode	16K	4	Valid	343.2	86.1	86 ms	11.6 ms	4.0	0	10,296	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	98.9	99.1	49 ms	10.1 ms	1.0	0	2,966	30.0 s
Sustained decode	32K	2	Valid	146.2	73.3	69 ms	13.6 ms	2.0	0	4,386	30.0 s
Sustained decode	32K	4	Valid	239.9	60.3	108 ms	16.6 ms	4.0	0	7,196	30.0 s
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	71.1	71.2	120 ms	14.0 ms	1.0	0	2,134	30.0 s
Sustained decode	64K	2	Valid	95.6	47.8	182 ms	20.9 ms	2.0	0	2,868	30.0 s
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	128K	1	Valid	45.4	45.4	248 ms	22.0 ms	1.0	0	1,361	30.0 s
Sustained decode	128K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	8	Over capacity	—	—	—	—	—	—	—	—
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	160K	1	Valid	38.4	38.4	297 ms	26.1 ms	1.0	0	1,150	30.0 s
Sustained decode	160K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	160K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	160K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	160K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	160K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	160K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	160K	128	Over capacity	—	—	—	—	—	—	—	—

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