Speed test matrix

NVIDIA-Nemotron-Nano-9B-v2-quantized.w4a16 on RTX 3090 24GB

vllm · 0.32.0 · W4A165 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 05:17 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: W4A16
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-quantized.w4a16 --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 context4,9901,681 ms TTFT

16K context4,8553,421 ms TTFT

32K context4,6327,140 ms TTFT

64K context4,26315,475 ms TTFT

120K context3,81532,390 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	129.230 ms TTFT	245.6182 ms TTFT	446.585 ms TTFT	755.8149 ms TTFT	1,150.4287 ms TTFT	1,494.3456 ms TTFT	1,689.9889 ms TTFT	Over capacity
16K	124.2206 ms TTFT	229.6305 ms TTFT	411.8502 ms TTFT	653.9904 ms TTFT	907.81,735 ms TTFT	Over capacity	Over capacity	Over capacity
32K	119.7114 ms TTFT	215.7161 ms TTFT	381.0248 ms TTFT	572.0431 ms TTFT	730.7831 ms TTFT	Over capacity	Over capacity	Over capacity
64K	111.6186 ms TTFT	192.2251 ms TTFT	330.7372 ms TTFT	459.2640 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
120K	99.7250 ms TTFT	161.2338 ms TTFT	267.9533 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	4,990.0	—	1,681 ms	—	—	1	—	—
Prefill	16K	1	Valid	4,855.0	—	3,421 ms	—	—	1	—	—
Prefill	32K	1	Valid	4,632.0	—	7,140 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,263.0	—	15,475 ms	—	—	1	—	—
Prefill	120K	1	Valid	3,815.0	—	32,390 ms	—	—	1	—	—
Sustained decode	0	1	Valid	129.2	129.4	30 ms	7.7 ms	1.0	0	3,877	30.0 s
Sustained decode	0	2	Valid	245.6	123.1	182 ms	8.1 ms	2.0	0	7,366	30.0 s
Sustained decode	0	4	Valid	446.5	111.9	85 ms	8.9 ms	4.0	0	13,392	30.0 s
Sustained decode	0	8	Valid	755.8	94.9	149 ms	10.5 ms	8.0	0	22,672	30.0 s
Sustained decode	0	16	Valid	1,150.4	72.2	287 ms	13.9 ms	16.0	0	34,512	30.0 s
Sustained decode	0	32	Valid	1,494.3	46.8	456 ms	21.4 ms	32.0	0	44,800	30.0 s
Sustained decode	0	64	Valid	1,689.9	26.3	889 ms	38.0 ms	64.0	0	50,688	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	124.2	124.3	206 ms	8.0 ms	1.0	0	3,725	30.0 s
Sustained decode	16K	2	Valid	229.6	114.6	305 ms	8.7 ms	2.0	0	6,889	30.0 s
Sustained decode	16K	4	Valid	411.8	102.3	502 ms	9.8 ms	4.0	0	12,352	30.0 s
Sustained decode	16K	8	Valid	653.9	80.3	904 ms	12.5 ms	8.0	0	19,616	30.0 s
Sustained decode	16K	16	Valid	907.8	54.4	1,735 ms	18.4 ms	16.0	0	27,232	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	119.7	119.8	114 ms	8.3 ms	1.0	0	3,591	30.0 s
Sustained decode	32K	2	Valid	215.7	107.9	161 ms	9.3 ms	2.0	0	6,470	30.0 s
Sustained decode	32K	4	Valid	381.0	95.1	248 ms	10.5 ms	4.0	0	11,428	30.0 s
Sustained decode	32K	8	Valid	572.0	71.2	431 ms	14.0 ms	8.0	0	17,160	30.0 s
Sustained decode	32K	16	Valid	730.7	45.4	831 ms	22.0 ms	16.0	0	21,920	30.0 s
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	111.6	111.7	186 ms	9.0 ms	1.0	0	3,347	30.0 s
Sustained decode	64K	2	Valid	192.2	96.1	251 ms	10.4 ms	2.0	0	5,766	30.0 s
Sustained decode	64K	4	Valid	330.7	82.4	372 ms	12.1 ms	4.0	0	9,921	30.0 s
Sustained decode	64K	8	Valid	459.2	56.6	640 ms	17.7 ms	8.0	0	13,768	30.0 s
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	99.7	99.8	250 ms	10.0 ms	1.0	0	2,991	30.0 s
Sustained decode	120K	2	Valid	161.2	80.7	338 ms	12.4 ms	2.0	0	4,836	30.0 s
Sustained decode	120K	4	Valid	267.9	66.8	533 ms	15.0 ms	4.0	0	8,036	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.