Speed test matrix

NVIDIA-Nemotron-Nano-9B-v2-FP8-dynamic on RTX 3090 24GB

vllm · 0.32.0 · FP8 dynamic5 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 15:06 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 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), 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 RedHatAI/NVIDIA-Nemotron-Nano-9B-v2-FP8-dynamic --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context5,7571,457 ms TTFT

16K context5,6622,934 ms TTFT

32K context5,4246,098 ms TTFT

64K context4,99413,212 ms TTFT

120K context4,38928,155 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	84.629 ms TTFT	163.146 ms TTFT	305.779 ms TTFT	543.5141 ms TTFT	896.1202 ms TTFT	1,317.6385 ms TTFT	1,682.1770 ms TTFT	Over capacity
16K	82.3183 ms TTFT	155.8272 ms TTFT	289.5443 ms TTFT	489.5800 ms TTFT	741.01,548 ms TTFT	Over capacity	Over capacity	Over capacity
32K	80.3122 ms TTFT	149.3154 ms TTFT	274.2241 ms TTFT	442.1422 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	76.5177 ms TTFT	137.6237 ms TTFT	248.1359 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	70.7248 ms TTFT	121.1364 ms TTFT	212.4532 ms TTFT	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	5,757.0	—	1,457 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,662.0	—	2,934 ms	—	—	1	—	—
Prefill	32K	1	Valid	5,424.0	—	6,098 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,994.0	—	13,212 ms	—	—	1	—	—
Prefill	120K	1	Valid	4,389.0	—	28,155 ms	—	—	1	—	—
Sustained decode	0	1	Valid	84.6	84.6	29 ms	11.8 ms	1.0	0	2,536	30.0 s
Sustained decode	0	2	Valid	163.1	81.6	46 ms	12.3 ms	2.0	0	4,892	30.0 s
Sustained decode	0	4	Valid	305.7	76.5	79 ms	13.1 ms	4.0	0	9,168	30.0 s
Sustained decode	0	8	Valid	543.5	68.1	141 ms	14.7 ms	8.0	0	16,304	30.0 s
Sustained decode	0	16	Valid	896.1	56.1	202 ms	17.8 ms	16.0	0	26,880	30.0 s
Sustained decode	0	32	Valid	1,317.6	41.2	385 ms	24.3 ms	32.0	0	39,520	30.0 s
Sustained decode	0	64	Valid	1,682.1	26.2	770 ms	38.1 ms	64.0	0	50,432	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	82.3	82.3	183 ms	12.1 ms	1.0	0	2,470	30.0 s
Sustained decode	16K	2	Valid	155.8	77.8	272 ms	12.9 ms	2.0	0	4,674	30.0 s
Sustained decode	16K	4	Valid	289.5	71.9	443 ms	13.9 ms	4.0	0	8,684	30.0 s
Sustained decode	16K	8	Valid	489.5	60.2	800 ms	16.6 ms	8.0	0	14,680	30.0 s
Sustained decode	16K	16	Valid	741.0	44.7	1,548 ms	22.4 ms	16.0	0	22,224	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	80.3	80.3	122 ms	12.4 ms	1.0	0	2,409	30.0 s
Sustained decode	32K	2	Valid	149.3	74.6	154 ms	13.4 ms	2.0	0	4,478	30.0 s
Sustained decode	32K	4	Valid	274.2	68.5	241 ms	14.6 ms	4.0	0	8,224	30.0 s
Sustained decode	32K	8	Valid	442.1	55.1	422 ms	18.1 ms	8.0	0	13,256	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	76.5	76.6	177 ms	13.1 ms	1.0	0	2,296	30.0 s
Sustained decode	64K	2	Valid	137.6	68.8	237 ms	14.5 ms	2.0	0	4,126	30.0 s
Sustained decode	64K	4	Valid	248.1	61.8	359 ms	16.2 ms	4.0	0	7,440	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	120K	1	Valid	70.7	70.8	248 ms	14.1 ms	1.0	0	2,122	30.0 s
Sustained decode	120K	2	Valid	121.1	60.5	364 ms	16.5 ms	2.0	0	3,630	30.0 s
Sustained decode	120K	4	Valid	212.4	52.9	532 ms	18.9 ms	4.0	0	6,372	30.0 s
Sustained decode	120K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	128	Over capacity	—	—	—	—	—	—	—	—

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