Best AI models for a Mac with 64 GB of memory
Apple Silicon Macs share one pool of memory between the processor and the GPU, so a Mac with 64 GB can hold models that would need several graphics cards on a PC. That makes Macs one of the most popular ways to run large models at home.
The table lists the most downloaded models that fit on a 64 GB Mac with at least 16K tokens of context. It's a popularity ranking of what fits, not a quality ranking. Each name opens the full answer for this Mac.
Nothing the site has analyzed fits with these settings.
0 models the site has analyzed fit on a Mac with 64 GB with at least 16K tokens of context. See them all in the hardware calculator.
How much of the memory can a model use?
Not all of it: macOS keeps some for itself. By default it lets the GPU use about three quarters of the memory on
Macs with more than 36 GB, so about 48 GB here, and that's what the table assumes. Advanced users can raise the
limit with the iogpu.wired_limit_mb system setting, at the risk of leaving too little for everything else.
Tips
- Mixture-of-experts models suit Macs well. They only use a fraction of their parameters for each token, so they run fast even though all of their experts sit in memory.
- MLX versions (repos with "MLX" in the name) are made for Apple Silicon and run through Apple's MLX library or LM Studio. GGUF files run in llama.cpp, Ollama and LM Studio.
- Other sizes: 32 GB, 128 GB, or any Mac in the hardware calculator.