Speed test matrix

MiniCPM5-1B 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 30, 2026, 21:05 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

# 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 openbmb/MiniCPM5-1B --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 context36.8K219 ms TTFT

16K context28.8K556 ms TTFT

32K context19.1K1,670 ms TTFT

64K context11.8K5,392 ms TTFT

120K context6,96317,092 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	298.112 ms TTFT	538.523 ms TTFT	981.523 ms TTFT	Looped	Looped	4,618.665 ms TTFT	Looped	Over capacity
16K	250.556 ms TTFT	441.380 ms TTFT	Looped	Looped	Looped	2,490.8605 ms TTFT	Over capacity	Over capacity
32K	220.8116 ms TTFT	376.6134 ms TTFT	Looped	955.5309 ms TTFT	1,323.1562 ms TTFT	Over capacity	Over capacity	Over capacity
64K	Looped	290.7295 ms TTFT	Looped	623.4626 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
120K	145.9428 ms TTFT	206.3527 ms TTFT	316.8739 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	36.8K	—	219 ms	—	—	1	—	—
Prefill	16K	1	Valid	28.8K	—	556 ms	—	—	1	—	—
Prefill	32K	1	Valid	19.1K	—	1,670 ms	—	—	1	—	—
Prefill	64K	1	Valid	11.8K	—	5,392 ms	—	—	1	—	—
Prefill	120K	1	Valid	6,963.0	—	17,092 ms	—	—	1	—	—
Sustained decode	0	1	Valid	298.1	301.7	12 ms	3.3 ms	1.0	1	8,942	30.0 s
Sustained decode	0	2	Valid	538.5	279.3	23 ms	3.6 ms	2.0	2	16,155	30.0 s
Sustained decode	0	4	Valid	981.5	264.1	23 ms	3.8 ms	4.0	4	29,444	30.0 s
Sustained decode	0	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	32	Valid	4,618.6	159.6	65 ms	6.3 ms	32.0	0	138,561	30.0 s
Sustained decode	0	64	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	250.5	262.0	56 ms	3.8 ms	1.0	1	7,515	30.0 s
Sustained decode	16K	2	Valid	441.3	222.0	80 ms	4.5 ms	2.0	0	13,240	30.0 s
Sustained decode	16K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	32	Valid	2,490.8	77.6	605 ms	12.9 ms	32.0	0	74,720	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	220.8	221.5	116 ms	4.5 ms	1.0	0	6,624	30.0 s
Sustained decode	32K	2	Valid	376.6	189.0	134 ms	5.3 ms	2.0	0	11,299	30.0 s
Sustained decode	32K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	8	Valid	955.5	120.0	309 ms	8.3 ms	8.0	0	28,664	30.0 s
Sustained decode	32K	16	Valid	1,323.1	82.2	562 ms	12.2 ms	16.0	0	39,696	30.0 s
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	2	Valid	290.7	145.7	295 ms	6.9 ms	2.0	0	8,720	30.0 s
Sustained decode	64K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	8	Valid	623.4	77.8	626 ms	12.9 ms	8.0	0	18,704	30.0 s
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	145.9	146.4	428 ms	6.8 ms	1.0	0	4,376	30.0 s
Sustained decode	120K	2	Valid	206.3	103.4	527 ms	9.7 ms	2.0	0	6,188	30.0 s
Sustained decode	120K	4	Valid	316.8	79.2	739 ms	12.6 ms	4.0	0	9,503	30.0 s
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.