RTX 3090 vs RTX 4090

For running local LLMs · 6 reports across 3 models

Side A

Vendor: nvidia
VRAM: 24GB
Memory: Discrete

Side B

Vendor: nvidia
VRAM: 24GB
Memory: Discrete

Tokens per second by model

Model	RTX 3090	RTX 4090
Qwen3.6up to 35B	66.0n=2	25.0n=1
Gemma 4up to 31B	—	149.6n=2
Qwen2.5	28.0n=1	—

More comparisons