New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
Hardware for local LLMs
GPUs and unified-memory systems ranked for running downloadable models — memory, bandwidth, max model, estimated tokens/sec.
Consumer Apple M4 Max 128GB
Consumer Apple M4 Max 64GB
Consumer Apple M4 Max 48GB
Consumer Apple M4 Max 36GB
Consumer Apple M3 Max 128GB
Consumer Apple M2 Max 96GB
Consumer Apple M1 Max 64GB
Consumer Apple M3 Max 64GB
Consumer Apple M2 Max 64GB
Consumer Apple M3 Max 48GB
Consumer Apple M1 Max 32GB
Consumer Apple M2 Max 32GB
Consumer Apple M3 Max 36GB
Prosumer Apple M2 Ultra 192GB
Prosumer Apple M1 Ultra 128GB
Prosumer Apple M2 Ultra 64GB
Prosumer Apple M1 Ultra 64GB
How to read this
Two numbers dominate local inference: memory (how big a model + context fits) and bandwidth (token speed). Click any hardware for the full size×quant matrix, max usable context and measured reports, or browse runtimes.