Best GPU for local LLMs, by budget
VRAM decides which models run; bandwidth decides how fast. Pick a budget to see every consumer GPU at or under that launch price, with how many of the 286 open models we track each runs at 4-bit or better, and what happens with popular models like gpt-oss and Qwen3.
Under $300
4 cards
Most VRAM: Arc B580 (12 GB) · runs 142 models at 4-bit+
Under $500
13 cards
Most VRAM: RX 7800 XT (16 GB) · runs 159 models at 4-bit+
Under $800
23 cards
Most VRAM: RTX 5070 Ti (16 GB) · runs 159 models at 4-bit+
Under $1,000
27 cards
Most VRAM: RX 7900 XTX (24 GB) · runs 182 models at 4-bit+
Under $1,500
29 cards
Most VRAM: RX 7900 XTX (24 GB) · runs 182 models at 4-bit+
Under $2,000
32 cards
Most VRAM: RTX 5090 (32 GB) · runs 200 models at 4-bit+
Already have a card? See models by VRAM size, the full GPU list, or Mac vs PC.
Launch MSRPs, not street prices. Every number comes from the formulas on the methodology page.