Speed test matrix

Phi-4-reasoning-plus-NVFP4 on RTX 3090 24GB

vllm · 0.32.0 · NVFP43 contexts (0–24K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 24K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 15:16 UTC

Hardware

Type: Discrete GPU
GPU: RTX 3090
VRAM: 24 GB
CPU: Intel Core i7-9700K (8C)
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: none. KV cache: int4_per_token_head
# 1x RTX 3090 24GB (Zagitta/gpu-unlock), 420 W limit, PCIe Gen3 x16 | Intel Core i7-9700K 8C | 32 GB DDR4 | Ubuntu 24.04.5, NVIDIA driver 580.178.04
vllm serve nvidia/Phi-4-reasoning-plus-NVFP4 --max-model-len 32768 --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 context2,3653,569 ms TTFT

16K context1,7169,623 ms TTFT

24K context1,35118,207 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	72.643 ms TTFT	140.868 ms TTFT	262.272 ms TTFT	468.177 ms TTFT	814.690 ms TTFT	Over capacity	Over capacity	Over capacity
16K	51.4102 ms TTFT	91.9150 ms TTFT	135.8216 ms TTFT	178.4299 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
24K	44.7144 ms TTFT	78.2202 ms TTFT	108.8290 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	2,365.0	—	3,569 ms	—	—	1	—	—
Prefill	16K	1	Valid	1,716.0	—	9,623 ms	—	—	1	—	—
Prefill	24K	1	Valid	1,351.0	—	18,207 ms	—	—	1	—	—
Sustained decode	0	1	Valid	72.6	73.0	43 ms	13.7 ms	1.0	0	2,176	30.0 s
Sustained decode	0	2	Valid	140.8	71.0	68 ms	14.1 ms	2.0	0	4,224	30.0 s
Sustained decode	0	4	Valid	262.2	66.4	72 ms	15.1 ms	4.0	0	7,864	30.0 s
Sustained decode	0	8	Valid	468.1	59.9	77 ms	16.7 ms	8.0	0	14,040	30.0 s
Sustained decode	0	16	Valid	814.6	52.8	90 ms	18.9 ms	16.0	0	24,432	30.0 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	51.4	51.5	102 ms	19.4 ms	1.0	0	1,540	30.0 s
Sustained decode	16K	2	Valid	91.9	46.1	150 ms	21.7 ms	2.0	0	2,758	30.0 s
Sustained decode	16K	4	Valid	135.8	34.1	216 ms	29.4 ms	4.0	0	4,075	30.0 s
Sustained decode	16K	8	Valid	178.4	22.3	299 ms	44.8 ms	8.0	0	5,352	30.0 s
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	24K	1	Valid	44.7	44.8	144 ms	22.3 ms	1.0	0	1,340	30.0 s
Sustained decode	24K	2	Valid	78.2	39.1	202 ms	25.5 ms	2.0	0	2,344	30.0 s
Sustained decode	24K	4	Valid	108.8	27.3	290 ms	36.7 ms	4.0	0	3,264	30.0 s
Sustained decode	24K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	24K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	24K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	24K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	24K	128	Over capacity	—	—	—	—	—	—	—	—

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