Speed test matrix

Qwen3-1.7B-FP8 on RTX 3090 24GB

vllm · 0.32.0 · FP83 contexts (0–32K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 20:05 UTC

Hardware

Type: Discrete GPU
GPU: RTX 3090
VRAM: 24 GB
CPU: AMD Ryzen 9 7900X (12C)
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), 350 W limit (VBIOS max), PCIe Gen4 x16 | AMD Ryzen 9 7900X 12C | 32 GB DDR5 | Ubuntu 24.04.5, NVIDIA driver 595.99.02
vllm serve Qwen/Qwen3-1.7B-FP8 --max-model-len 40960 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context20.1K411 ms TTFT

16K context16.9K969 ms TTFT

32K context12.2K2,658 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	Looped	394.315 ms TTFT	672.517 ms TTFT	1,068.024 ms TTFT	1,630.651 ms TTFT	Over capacity	Over capacity	Over capacity
16K	155.143 ms TTFT	233.757 ms TTFT	367.486 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
32K	117.878 ms TTFT	162.4106 ms TTFT	237.0147 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	20.1K	—	411 ms	—	—	1	—	—
Prefill	16K	1	Valid	16.9K	—	969 ms	—	—	1	—	—
Prefill	32K	1	Valid	12.2K	—	2,658 ms	—	—	1	—	—
Sustained decode	0	1	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	2	Valid	394.3	205.1	15 ms	4.9 ms	2.0	0	11,828	30.0 s
Sustained decode	0	4	Valid	672.5	178.6	17 ms	5.6 ms	4.0	0	20,172	30.0 s
Sustained decode	0	8	Valid	1,068.0	146.1	24 ms	6.8 ms	8.0	0	32,041	30.0 s
Sustained decode	0	16	Valid	1,630.6	114.7	51 ms	8.7 ms	16.0	0	48,912	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	155.1	156.1	43 ms	6.4 ms	1.0	0	4,654	30.0 s
Sustained decode	16K	2	Valid	233.7	117.9	57 ms	8.5 ms	2.0	0	7,010	30.0 s
Sustained decode	16K	4	Valid	367.4	92.9	86 ms	10.8 ms	4.0	0	11,020	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	117.8	118.2	78 ms	8.5 ms	1.0	0	3,534	30.0 s
Sustained decode	32K	2	Valid	162.4	81.4	106 ms	12.3 ms	2.0	0	4,870	30.0 s
Sustained decode	32K	4	Valid	237.0	59.4	147 ms	16.8 ms	4.0	0	7,108	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	—	—	—	—	—	—	—	—

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.