Speed test matrix

Qwen3.5-4B-FP8-dynamic on RTX 3060 12GB

vllm · 0.32.0 · FP8 dynamic6 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 05:12 UTC

Hardware

Type: Discrete GPU
GPU: RTX 3060
VRAM: 12 GB
CPU: Intel Xeon E5-2680 v4 (14C)
RAM: 32 GB
OS: Ubuntu 24.04.5

Engine

Engine: vllm
Version: 0.32.0
Quantization: FP8 dynamic
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. KV cache: int4_per_token_head
# 1x RTX 3060 12GB, stock 170 W limit, PCIe Gen3 x16 | Intel Xeon E5-2680 v4 14C | 32 GB DDR4-2400 | Ubuntu 24.04.5, NVIDIA driver 580.178.04
vllm serve RedHatAI/Qwen3.5-4B-FP8-dynamic --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context2,3703,438 ms TTFT

16K context1,9708,181 ms TTFT

32K context1,49621,451 ms TTFT

64K context1,00463,766 ms TTFT

128K context606.0211,148 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	49.782 ms TTFT	94.8136 ms TTFT	181.2234 ms TTFT	333.1387 ms TTFT	574.0636 ms TTFT	909.31,009 ms TTFT	Over capacity	Over capacity
16K	44.81,264 ms TTFT	80.31,879 ms TTFT	136.53,044 ms TTFT	214.85,403 ms TTFT	299.310,200 ms TTFT	Over capacity	Over capacity	Over capacity
32K	40.81,529 ms TTFT	69.92,254 ms TTFT	109.23,670 ms TTFT	157.86,505 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	34.61,291 ms TTFT	55.21,793 ms TTFT	78.22,807 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	24.0Requests queued3,467 ms TTFT	31.6Requests queued4,967 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

Requests queued: the server could not run all requests at once, so some waited in a queue. 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	2,370.0	—	3,438 ms	—	—	1	—	—
Prefill	16K	1	Valid	1,970.0	—	8,181 ms	—	—	1	—	—
Prefill	32K	1	Valid	1,496.0	—	21,451 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,004.0	—	63,766 ms	—	—	1	—	—
Prefill	128K	1	Valid	606.0	—	211,148 ms	—	—	1	—	—
Sustained decode	0	1	Valid	49.7	49.8	82 ms	20.1 ms	1.0	0	1,491	30.0 s
Sustained decode	0	2	Valid	94.8	47.5	136 ms	21.1 ms	2.0	0	2,844	30.0 s
Sustained decode	0	4	Valid	181.2	45.4	234 ms	22.0 ms	4.0	0	5,435	30.0 s
Sustained decode	0	8	Valid	333.1	41.8	387 ms	23.9 ms	8.0	0	9,992	30.0 s
Sustained decode	0	16	Valid	574.0	36.1	636 ms	27.7 ms	16.0	0	17,216	30.0 s
Sustained decode	0	32	Valid	909.3	28.6	1,009 ms	35.0 ms	32.0	0	27,264	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	44.8	44.8	1,264 ms	22.3 ms	1.0	0	1,343	30.0 s
Sustained decode	16K	2	Valid	80.3	39.5	1,879 ms	25.3 ms	2.0	0	2,408	30.0 s
Sustained decode	16K	4	Valid	136.5	32.6	3,044 ms	30.7 ms	4.0	0	4,095	30.0 s
Sustained decode	16K	8	Valid	214.8	24.1	5,403 ms	41.5 ms	8.0	0	6,440	30.0 s
Sustained decode	16K	16	Valid	299.3	15.0	10,200 ms	66.8 ms	16.0	0	8,976	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	40.8	40.8	1,529 ms	24.5 ms	1.0	0	1,223	30.0 s
Sustained decode	32K	2	Valid	69.9	34.3	2,254 ms	29.1 ms	2.0	0	2,094	30.0 s
Sustained decode	32K	4	Valid	109.2	25.8	3,670 ms	38.7 ms	4.0	0	3,272	30.0 s
Sustained decode	32K	8	Valid	157.8	17.4	6,505 ms	57.5 ms	8.0	0	4,728	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	34.6	34.7	1,291 ms	28.9 ms	1.0	0	1,036	30.0 s
Sustained decode	64K	2	Valid	55.2	27.3	1,793 ms	36.7 ms	2.0	0	1,654	30.0 s
Sustained decode	64K	4	Valid	78.2	18.8	2,807 ms	53.1 ms	4.0	0	2,344	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	Requests queued	24.0	26.6	3,467 ms	37.6 ms	0.9	0	720	30.0 s
Sustained decode	128K	2	Requests queued	31.6	18.6	4,967 ms	54.0 ms	1.8	0	946	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	Over capacity	—	—	—	—	0.7	0	0	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	—	—	—	—	—	—	—	—

Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.