Speed test matrix

NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 on RTX 3090 24GB

vllm · 0.32.0 · NVFP44 contexts (0–56K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 56K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 19:39 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: NVFP4
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: MTP, 3 tokens lookahead
# 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 nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4 --max-model-len 65536 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code  --speculative-config {"method":"mtp","num_speculative_tokens":3} --max-cudagraph-capture-size 16

Prefill

prompt tok/s by context length

8K context8,1511,030 ms TTFT

16K context8,0172,073 ms TTFT

32K context7,3434,505 ms TTFT

56K context6,5218,857 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	208.9399 ms TTFT	335.01,213 ms TTFT	364.7Requests queued23,965 ms TTFT	333.6Requests queued24,725 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
16K	211.41,065 ms TTFT	207.8Requests queued40,457 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
32K	212.91,201 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
56K	198.81,328 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. 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	8,151.0	—	1,030 ms	—	—	1	—	—
Prefill	16K	1	Valid	8,017.0	—	2,073 ms	—	—	1	—	—
Prefill	32K	1	Valid	7,343.0	—	4,505 ms	—	—	1	—	—
Prefill	56K	1	Valid	6,521.0	—	8,857 ms	—	—	1	—	—
Sustained decode	0	1	Valid	208.9	208.8	399 ms	4.8 ms	1.0	0	6,267	30.0 s
Sustained decode	0	2	Valid	335.0	168.9	1,213 ms	5.9 ms	2.0	0	10,047	30.0 s
Sustained decode	0	4	Requests queued	364.7	169.0	23,965 ms	5.9 ms	2.0	2	10,939	30.0 s
Sustained decode	0	8	Requests queued	333.6	166.6	24,725 ms	6.0 ms	2.0	2	10,005	30.0 s
Sustained decode	0	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	211.4	212.5	1,065 ms	4.7 ms	1.0	0	6,340	30.0 s
Sustained decode	16K	2	Requests queued	207.8	213.9	40,457 ms	4.7 ms	1.0	2	6,232	30.0 s
Sustained decode	16K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	8	Over capacity	—	—	—	—	—	—	—	—
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	212.9	213.5	1,201 ms	4.7 ms	1.0	0	6,385	30.0 s
Sustained decode	32K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	4	Over capacity	—	—	—	—	—	—	—	—
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	56K	1	Valid	198.8	200.4	1,328 ms	5.0 ms	1.0	0	5,963	30.0 s
Sustained decode	56K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	56K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	56K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	56K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	56K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	56K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	56K	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.