Speed test matrix

Qwen3.5-4B on RTX 3090 24GB

vllm · 0.32.0 · BF166 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 22:23 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: MTP, 2 tokens lookahead
# 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 Qwen/Qwen3.5-4B --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2}

Prefill

prompt tok/s by context length

8K context11.1K735 ms TTFT

16K context10.5K1,546 ms TTFT

32K context9,0903,552 ms TTFT

64K context7,5278,558 ms TTFT

128K context5,42223,736 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	125.942 ms TTFT	237.848 ms TTFT	457.190 ms TTFT	817.9108 ms TTFT	1,388.2188 ms TTFT	2,039.0238 ms TTFT	Over capacity	Over capacity
16K	95.6140 ms TTFT	184.9213 ms TTFT	358.8370 ms TTFT	517.6607 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	71.7178 ms TTFT	146.3270 ms TTFT	275.2444 ms TTFT	366.8707 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	48.0316 ms TTFT	101.4452 ms TTFT	189.7659 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	32.9545 ms TTFT	62.9740 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	18.2937 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	11.1K	—	735 ms	—	—	1	—	—
Prefill	16K	1	Valid	10.5K	—	1,546 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,090.0	—	3,552 ms	—	—	1	—	—
Prefill	64K	1	Valid	7,527.0	—	8,558 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,422.0	—	23,736 ms	—	—	1	—	—
Sustained decode	0	1	Valid	125.9	127.5	42 ms	7.8 ms	1.0	0	3,775	30.0 s
Sustained decode	0	2	Valid	237.8	122.1	48 ms	8.2 ms	2.0	0	7,134	30.0 s
Sustained decode	0	4	Valid	457.1	117.0	90 ms	8.5 ms	4.0	0	13,712	30.0 s
Sustained decode	0	8	Valid	817.9	104.7	108 ms	9.6 ms	8.0	0	24,532	30.0 s
Sustained decode	0	16	Valid	1,388.2	89.0	188 ms	11.2 ms	16.0	0	41,625	30.0 s
Sustained decode	0	32	Valid	2,039.0	65.9	238 ms	15.2 ms	32.0	0	61,170	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	95.6	96.9	140 ms	10.3 ms	1.0	0	2,866	30.0 s
Sustained decode	16K	2	Valid	184.9	93.5	213 ms	10.7 ms	2.0	0	5,544	30.0 s
Sustained decode	16K	4	Valid	358.8	90.5	370 ms	11.0 ms	4.0	0	10,763	30.0 s
Sustained decode	16K	8	Valid	517.6	65.6	607 ms	15.3 ms	8.0	0	15,530	30.0 s
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	71.7	73.0	178 ms	13.7 ms	1.0	0	2,148	30.0 s
Sustained decode	32K	2	Valid	146.3	74.1	270 ms	13.5 ms	2.0	0	4,385	30.0 s
Sustained decode	32K	4	Valid	275.2	69.9	444 ms	14.3 ms	4.0	0	8,255	30.0 s
Sustained decode	32K	8	Valid	366.8	46.9	707 ms	21.3 ms	8.0	0	10,989	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	48.0	49.5	316 ms	20.2 ms	1.0	0	1,441	30.0 s
Sustained decode	64K	2	Valid	101.4	51.6	452 ms	19.4 ms	2.0	0	3,039	30.0 s
Sustained decode	64K	4	Valid	189.7	48.6	659 ms	20.6 ms	4.0	0	5,689	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	128K	1	Valid	32.9	33.3	545 ms	30.0 ms	1.0	0	986	30.0 s
Sustained decode	128K	2	Valid	62.9	31.9	740 ms	31.3 ms	2.0	0	1,885	30.0 s
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	248K	1	Valid	18.2	18.9	937 ms	52.9 ms	1.0	0	547	30.0 s
Sustained decode	248K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	4	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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