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, 13:31 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. KV cache: int4_per_token_head
# 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 --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context3,5872,338 ms TTFT

16K context3,1745,234 ms TTFT

32K context2,52513,099 ms TTFT

64K context1,82436,167 ms TTFT

120K context1,228100,611 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	94.238 ms TTFT	180.563 ms TTFT	333.4105 ms TTFT	580.9183 ms TTFT	916.3270 ms TTFT	1,286.5560 ms TTFT	1,469.41,137 ms TTFT	1,214.7Requests queued1,024 ms TTFT
16K	86.2753 ms TTFT	154.81,130 ms TTFT	268.41,887 ms TTFT	427.53,384 ms TTFT	603.16,413 ms TTFT	766.012,548 ms TTFT	Over capacity	Over capacity
32K	79.82,000 ms TTFT	135.72,970 ms TTFT	224.84,933 ms TTFT	338.68,816 ms TTFT	451.416,760 ms TTFT	Over capacity	Over capacity	Over capacity
64K	69.32,045 ms TTFT	109.93,017 ms TTFT	170.64,964 ms TTFT	240.38,954 ms TTFT	300.516,777 ms TTFT	Over capacity	Over capacity	Over capacity
120K	56.71,768 ms TTFT	82.22,603 ms TTFT	120.94,267 ms TTFT	160.27,670 ms TTFT	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. 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	3,587.0	—	2,338 ms	—	—	1	—	—
Prefill	16K	1	Valid	3,174.0	—	5,234 ms	—	—	1	—	—
Prefill	32K	1	Valid	2,525.0	—	13,099 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,824.0	—	36,167 ms	—	—	1	—	—
Prefill	120K	1	Valid	1,228.0	—	100,611 ms	—	—	1	—	—
Sustained decode	0	1	Valid	94.2	94.3	38 ms	10.6 ms	1.0	0	2,824	30.0 s
Sustained decode	0	2	Valid	180.5	90.4	63 ms	11.1 ms	2.0	0	5,414	30.0 s
Sustained decode	0	4	Valid	333.4	83.6	105 ms	12.0 ms	4.0	0	10,000	30.0 s
Sustained decode	0	8	Valid	580.9	72.9	183 ms	13.7 ms	8.0	0	17,424	30.0 s
Sustained decode	0	16	Valid	916.3	57.5	270 ms	17.4 ms	16.0	0	27,491	30.0 s
Sustained decode	0	32	Valid	1,286.5	40.3	560 ms	24.8 ms	32.0	0	38,592	30.0 s
Sustained decode	0	64	Valid	1,469.4	22.9	1,137 ms	43.7 ms	64.0	0	44,032	30.0 s
Sustained decode	0	128	Requests queued	1,214.7	18.4	1,024 ms	54.3 ms	67.0	0	36,381	29.9 s
Sustained decode	16K	1	Valid	86.2	86.2	753 ms	11.6 ms	1.0	0	2,584	30.0 s
Sustained decode	16K	2	Valid	154.8	76.7	1,130 ms	13.0 ms	2.0	0	4,642	30.0 s
Sustained decode	16K	4	Valid	268.4	65.1	1,887 ms	15.4 ms	4.0	0	8,052	30.0 s
Sustained decode	16K	8	Valid	427.5	49.8	3,384 ms	20.1 ms	8.0	0	12,824	30.0 s
Sustained decode	16K	16	Valid	603.1	32.6	6,413 ms	30.7 ms	16.0	0	18,080	30.0 s
Sustained decode	16K	32	Valid	766.0	18.2	12,548 ms	54.8 ms	32.0	0	22,976	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	79.8	79.8	2,000 ms	12.5 ms	1.0	0	2,392	30.0 s
Sustained decode	32K	2	Valid	135.7	66.1	2,970 ms	15.1 ms	2.0	0	4,070	30.0 s
Sustained decode	32K	4	Valid	224.8	51.8	4,933 ms	19.3 ms	4.0	0	6,744	30.0 s
Sustained decode	32K	8	Valid	338.6	35.4	8,816 ms	28.3 ms	8.0	0	10,152	30.0 s
Sustained decode	32K	16	Valid	451.4	19.8	16,760 ms	50.6 ms	16.0	0	13,536	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	69.3	69.4	2,045 ms	14.4 ms	1.0	0	2,080	30.0 s
Sustained decode	64K	2	Valid	109.9	53.5	3,017 ms	18.7 ms	2.0	0	3,294	30.0 s
Sustained decode	64K	4	Valid	170.6	39.4	4,964 ms	25.4 ms	4.0	0	5,116	30.0 s
Sustained decode	64K	8	Valid	240.3	25.1	8,954 ms	39.9 ms	8.0	0	7,208	30.0 s
Sustained decode	64K	16	Valid	300.5	13.3	16,777 ms	75.0 ms	16.0	0	9,008	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	1	Valid	56.7	56.8	1,768 ms	17.6 ms	1.0	0	1,701	30.0 s
Sustained decode	120K	2	Valid	82.2	40.2	2,603 ms	24.9 ms	2.0	0	2,466	30.0 s
Sustained decode	120K	4	Valid	120.9	28.3	4,267 ms	35.4 ms	4.0	0	3,624	30.0 s
Sustained decode	120K	8	Valid	160.2	17.2	7,670 ms	58.1 ms	8.0	0	4,800	30.0 s
Sustained decode	120K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	128	Over capacity	—	—	—	—	—	—	—	—

Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.