Honest benchmarks for running open-weights AI models on your own GPU. 84 models × 27 consumer GPUs — from RTX 3060 to Apple Silicon — each pair gets an honest verdict: runs, tight fit, or won't fit, with real speed and VRAM numbers. Plus 800+ step-by-step install recipes (llama.cpp, Ollama, ComfyUI, MLX) and a GPU Advisor. No invented numbers, no signup, free under CC BY-SA.
