Skip to main content
Local AI Rigs

GPU × Model Fit Matrix

Will a given GPU run a given local model? Each page below is a VRAM planning estimate — parameter count × bytes/parameter × quantization + context overhead — using the same math as our VRAM calculator. No invented benchmark numbers.

🟢 fits · 🟡 only at 4-bit (tight) · 🔴 needs more VRAM

Intel Arc B580

12GB · 7B to 8B models, with some 13B experimentation.

RTX 5060 Ti 16GB

16GB · 7B to 14B models, with 30B planning at aggressive quantization.

RTX 3090 (used)

24GB · 30B to 34B models, with 70B planning at aggressive quantization.

RTX 4090

24GB · 30B to 34B models, with 70B work possible but constrained.

AMD RX 7900 XTX

24GB · 30B-class planning when the software stack cooperates.

RTX 5090

32GB · 30B to 70B planning with the most room for context and quality.

Mac mini M4 Pro (64GB)

64GB · 7B to 32B local models when you value efficiency, silence, and unified memory over peak throughput.

Radeon Pro W7900

48GB · 70B-class models with more room for longer contexts and lighter batching.

RTX A6000 (used)

48GB · 70B-class local models and more serious context-heavy inference planning.

RTX 6000 Ada

48GB · 70B-class deployment with more headroom for heavier precision and longer sessions.

RTX PRO 6000 Blackwell

96GB · 120B-class single-GPU planning and much more breathing room for long contexts.

Mac Studio M4 Max (128GB)

128GB · 70B-class planning with far better efficiency than large discrete-GPU towers.