Speed test matrix

Qwen3.5-4B-AWQ-4bit on RTX 3090 24GB

vllm · 0.32.0 · AWQ 4-bit6 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 07:26 UTCEdited Oct 29, 2026, 12:41 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: AWQ 4-bit
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 cyankiwi/Qwen3.5-4B-AWQ-4bit --max-model-len 262144 --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 context11.8K695 ms TTFT

16K context11K1,474 ms TTFT

32K context9,8983,263 ms TTFT

64K context8,2467,812 ms TTFT

128K context6,18520,808 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	186.233 ms TTFT	338.937 ms TTFT	633.666 ms TTFT	1,096.293 ms TTFT	1,794.2153 ms TTFT	2,627.4244 ms TTFT	3,391.0485 ms TTFT	Over capacity
16K	167.698 ms TTFT	287.4128 ms TTFT	522.2184 ms TTFT	856.5300 ms TTFT	1,226.4578 ms TTFT	Over capacity	Over capacity	Over capacity
32K	152.7123 ms TTFT	248.8145 ms TTFT	447.2228 ms TTFT	682.1350 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	129.5187 ms TTFT	195.9253 ms TTFT	339.8374 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	99.2497 ms TTFT	137.1600 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	68.8888 ms TTFT	87.61,135 ms TTFT	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	11.8K	—	695 ms	—	—	1	—	—
Prefill	16K	1	Valid	11K	—	1,474 ms	—	—	1	—	—
Prefill	32K	1	Valid	9,898.0	—	3,263 ms	—	—	1	—	—
Prefill	64K	1	Valid	8,246.0	—	7,812 ms	—	—	1	—	—
Prefill	128K	1	Valid	6,185.0	—	20,808 ms	—	—	1	—	—
Sustained decode	0	1	Valid	186.2	187.1	33 ms	5.3 ms	1.0	0	5,587	30.0 s
Sustained decode	0	2	Valid	338.9	170.6	37 ms	5.9 ms	2.0	0	10,166	30.0 s
Sustained decode	0	4	Valid	633.6	160.2	66 ms	6.2 ms	4.0	0	19,004	30.0 s
Sustained decode	0	8	Valid	1,096.2	139.5	93 ms	7.2 ms	8.0	0	32,880	30.0 s
Sustained decode	0	16	Valid	1,794.2	114.8	153 ms	8.7 ms	16.0	0	53,824	30.0 s
Sustained decode	0	32	Valid	2,627.4	84.1	244 ms	11.9 ms	32.0	0	78,816	30.0 s
Sustained decode	0	64	Valid	3,391.0	54.1	485 ms	18.5 ms	64.0	0	101,696	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	167.6	168.0	98 ms	6.0 ms	1.0	0	5,029	30.0 s
Sustained decode	16K	2	Valid	287.4	144.0	128 ms	6.9 ms	2.0	0	8,620	30.0 s
Sustained decode	16K	4	Valid	522.2	130.9	184 ms	7.6 ms	4.0	0	15,663	30.0 s
Sustained decode	16K	8	Valid	856.5	107.0	300 ms	9.3 ms	8.0	0	25,688	30.0 s
Sustained decode	16K	16	Valid	1,226.4	76.1	578 ms	13.1 ms	16.0	0	36,784	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	152.7	153.0	123 ms	6.5 ms	1.0	0	4,579	30.0 s
Sustained decode	32K	2	Valid	248.8	124.7	145 ms	8.0 ms	2.0	0	7,463	30.0 s
Sustained decode	32K	4	Valid	447.2	112.0	228 ms	8.9 ms	4.0	0	13,416	30.0 s
Sustained decode	32K	8	Valid	682.1	85.3	350 ms	11.7 ms	8.0	0	20,464	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	129.5	129.7	187 ms	7.7 ms	1.0	0	3,884	30.0 s
Sustained decode	64K	2	Valid	195.9	98.1	253 ms	10.2 ms	2.0	0	5,877	30.0 s
Sustained decode	64K	4	Valid	339.8	85.0	374 ms	11.8 ms	4.0	0	10,195	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	99.2	99.5	497 ms	10.1 ms	1.0	0	2,975	30.0 s
Sustained decode	128K	2	Valid	137.1	68.6	600 ms	14.6 ms	2.0	0	4,112	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	68.8	69.0	888 ms	14.5 ms	1.0	0	2,062	30.0 s
Sustained decode	248K	2	Valid	87.6	43.8	1,135 ms	22.8 ms	2.0	0	2,626	30.0 s
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.