MacforAIBlog
Practical guides on choosing and using a Mac for local LLMs.
Why Memory Bandwidth Decides Tokens per Second on a Mac
Generation speed is limited by memory bandwidth, prompt reading by GPU compute. How to estimate tokens per second and why Pro, Max and Ultra chips feel so different.
How Much RAM Do You Need for a Local LLM on a Mac?
Weights, KV cache and the GPU memory limit explained: how to tell whether a local LLM fits your Mac, with rough RAM targets from 16 GB to 128 GB.