Speed test matrix

gemma-4-12B-it-FP8-Dynamic on RTX 3090 24GB

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 30, 2026, 03:55 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: 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
# 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 RedHatAI/gemma-4-12B-it-FP8-Dynamic --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 context3,8472,083 ms TTFT

16K context3,2194,932 ms TTFT

32K context2,54512,417 ms TTFT

64K context1,79735,087 ms TTFT

128K context1,125111,928 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	57.134 ms TTFT	111.239 ms TTFT	210.950 ms TTFT	384.360 ms TTFT	657.499 ms TTFT	935.4Requests queued161 ms TTFT	Over capacity	Over capacity
16K	50.595 ms TTFT	99.9125 ms TTFT	192.5170 ms TTFT	Looped	Over capacity	Over capacity	Over capacity	Over capacity
32K	47.1162 ms TTFT	Looped	Looped	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	Looped	Looped	Looped	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	37.4651 ms TTFT	74.3896 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	29.1Requests queued1,392 ms TTFT	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. 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	3,847.0	—	2,083 ms	—	—	1	—	—
Prefill	16K	1	Valid	3,219.0	—	4,932 ms	—	—	1	—	—
Prefill	32K	1	Valid	2,545.0	—	12,417 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,797.0	—	35,087 ms	—	—	1	—	—
Prefill	128K	1	Valid	1,125.0	—	111,928 ms	—	—	1	—	—
Sustained decode	0	1	Valid	57.1	57.4	34 ms	17.4 ms	1.0	0	1,713	30.0 s
Sustained decode	0	2	Valid	111.2	56.0	39 ms	17.9 ms	2.0	0	3,336	30.0 s
Sustained decode	0	4	Valid	210.9	53.3	50 ms	18.8 ms	4.0	0	6,327	30.0 s
Sustained decode	0	8	Valid	384.3	49.0	60 ms	20.4 ms	8.0	0	11,528	30.0 s
Sustained decode	0	16	Valid	657.4	42.5	99 ms	23.5 ms	16.0	0	19,712	30.0 s
Sustained decode	0	32	Requests queued	935.4	37.1	161 ms	26.9 ms	26.2	0	28,062	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	50.5	50.6	95 ms	19.8 ms	1.0	0	1,513	30.0 s
Sustained decode	16K	2	Valid	99.9	50.1	125 ms	20.0 ms	2.0	0	2,996	30.0 s
Sustained decode	16K	4	Valid	192.5	48.3	170 ms	20.7 ms	4.0	0	5,772	30.0 s
Sustained decode	16K	8	Looped	—	—	—	—	—	0	0	—
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	47.1	47.2	162 ms	21.2 ms	1.0	0	1,413	30.0 s
Sustained decode	32K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	4	Looped	—	—	—	—	—	0	0	—
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	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	4	Looped	—	—	—	—	—	0	0	—
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	37.4	37.5	651 ms	26.6 ms	1.0	0	1,122	30.0 s
Sustained decode	128K	2	Valid	74.3	37.2	896 ms	26.9 ms	2.0	0	2,229	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	Requests queued	29.1	30.0	1,392 ms	33.3 ms	1.0	0	873	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	—	—	—	—	—	—	—	—