Speed test matrix

NVIDIA-Nemotron-Nano-12B-v2-AWQ-4bit on RTX 3090 24GB

vllm · 0.32.0 · AWQ 4-bit5 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, 06:18 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: AWQ 4-bit
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 cyankiwi/NVIDIA-Nemotron-Nano-12B-v2-AWQ-4bit --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 context4,0102,092 ms TTFT

16K context3,8944,267 ms TTFT

32K context3,7098,917 ms TTFT

64K context3,37919,526 ms TTFT

120K context2,95541,818 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	96.437 ms TTFT	185.059 ms TTFT	342.8104 ms TTFT	598.3181 ms TTFT	950.0266 ms TTFT	1,310.2547 ms TTFT	Over capacity	Over capacity
16K	92.1254 ms TTFT	171.5377 ms TTFT	312.6626 ms TTFT	503.91,141 ms TTFT	705.52,199 ms TTFT	Over capacity	Over capacity	Over capacity
32K	88.4135 ms TTFT	159.9189 ms TTFT	286.1303 ms TTFT	431.5540 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	81.8215 ms TTFT	140.7291 ms TTFT	244.2460 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	72.4291 ms TTFT	116.5372 ms TTFT	Over capacity	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	4,010.0	—	2,092 ms	—	—	1	—	—
Prefill	16K	1	Valid	3,894.0	—	4,267 ms	—	—	1	—	—
Prefill	32K	1	Valid	3,709.0	—	8,917 ms	—	—	1	—	—
Prefill	64K	1	Valid	3,379.0	—	19,526 ms	—	—	1	—	—
Prefill	120K	1	Valid	2,955.0	—	41,818 ms	—	—	1	—	—
Sustained decode	0	1	Valid	96.4	96.5	37 ms	10.4 ms	1.0	0	2,892	30.0 s
Sustained decode	0	2	Valid	185.0	92.7	59 ms	10.8 ms	2.0	0	5,550	30.0 s
Sustained decode	0	4	Valid	342.8	85.9	104 ms	11.6 ms	4.0	0	10,284	30.0 s
Sustained decode	0	8	Valid	598.3	75.1	181 ms	13.3 ms	8.0	0	17,944	30.0 s
Sustained decode	0	16	Valid	950.0	59.5	266 ms	16.8 ms	16.0	0	28,496	30.0 s
Sustained decode	0	32	Valid	1,310.2	41.0	547 ms	24.4 ms	32.0	0	39,296	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	92.1	92.2	254 ms	10.8 ms	1.0	0	2,763	30.0 s
Sustained decode	16K	2	Valid	171.5	85.5	377 ms	11.7 ms	2.0	0	5,144	30.0 s
Sustained decode	16K	4	Valid	312.6	77.4	626 ms	12.9 ms	4.0	0	9,376	30.0 s
Sustained decode	16K	8	Valid	503.9	61.5	1,141 ms	16.3 ms	8.0	0	15,112	30.0 s
Sustained decode	16K	16	Valid	705.5	41.9	2,199 ms	23.9 ms	16.0	0	21,152	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	88.4	88.5	135 ms	11.3 ms	1.0	0	2,652	30.0 s
Sustained decode	32K	2	Valid	159.9	79.9	189 ms	12.5 ms	2.0	0	4,794	30.0 s
Sustained decode	32K	4	Valid	286.1	71.4	303 ms	14.0 ms	4.0	0	8,580	30.0 s
Sustained decode	32K	8	Valid	431.5	53.7	540 ms	18.6 ms	8.0	0	12,943	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	81.8	81.9	215 ms	12.2 ms	1.0	0	2,455	30.0 s
Sustained decode	64K	2	Valid	140.7	70.3	291 ms	14.2 ms	2.0	0	4,222	30.0 s
Sustained decode	64K	4	Valid	244.2	60.7	460 ms	16.5 ms	4.0	0	7,324	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	72.4	72.5	291 ms	13.8 ms	1.0	0	2,171	30.0 s
Sustained decode	120K	2	Valid	116.5	58.2	372 ms	17.2 ms	2.0	0	3,494	30.0 s
Sustained decode	120K	4	Over capacity	—	—	—	—	—	—	—	—
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.