Skip to content
AI Model Radar

Hardware

Buy memory,not speed.

Which machine runs which models is arithmetic, not opinion: measured file sizes, computed context cache, stated margins. Start with the guide that fits your budget, or jump straight to your card.

Guide · start here

Local AI on a budget

Which used and entry-level cards genuinely run local models, what to avoid, and what to check before buying second-hand. Fit counts computed, no street prices.

Read the guide →

Profile · the middle path

DGX Spark

128 GB of unified memory on a desk — capacity, not speed. When a personal AI box is rational, and when renting beats it.

Read the profile →

Every tracked GPU

EstimatedEstimated: computed from our curated model and hardware catalog — not a live reading.

Comfortable = rated Excellent or Good at 8K context by the same fit engine as every page here. Each card has its own page with every verdict and the arithmetic behind it.

Apple Silicon

Planned at 70% of unified memory — macOS caps what the GPU may take, and the pages say so.

Not sure what you have? The finder detects your GPU in the browser and gives every model a verdict against it.

Open the finder →