RTX 4090 vs RTX 5090

For running local LLMs · 16 reports across 4 models

Side A

Vendor: nvidia
VRAM: 24GB
Memory: Discrete

Side B

Vendor: nvidia
VRAM: 32GB
Memory: Discrete

Tokens per second by model

Model	RTX 4090	RTX 5090
Qwen3.6up to 35B	80.0n=2	3238.0n=7
Gemma 4up to 31B	149.6n=2	578.0n=3
Kimi K2.6	—	—n=1
Qwen2.5	—	—n=1

More comparisons