How to run an LLM locally on a Mac
A Mac keeps one pool of memory for the CPU and the GPU. That helps, because a model can use most of it. It also means the number on the box is not the number a model gets.
1. Know how much memory a model can use
macOS lets the GPU use only part of the unified memory by default. These are the calculator's figures for common sizes, and the second column is what you get after raising the limit.
| Mac memory | Default GPU limit | After raising it |
|---|---|---|
| 16GB | 10.7GB | Nothing to raise |
| 32GB | 21.3GB | 24GB |
| 64GB | 48GB | 56GB |
| 128GB | 96GB | 120GB |
To raise it, run sudo sysctl iogpu.wired_limit_mb=57344 (that example is for a 64GB Mac). The setting resets when you restart, and leaving macOS too little memory can make the whole machine unstable. The LLM VRAM calculator does this sum for your exact Mac.
2. Pick a runtime
LM Studio gives you a window with a model browser and a chat box. Ollama runs from a terminal. MLX is Apple's own framework for its chips. Each runtime page shows the speeds people report on it.
3. Choose a model that fits
Pick the biggest model that fits under your GPU limit with some room left for the context. The best Mac for local LLMs page ranks each chip by the speed people report, and the local LLM picker shows what people run at each memory size. If a quant name is unfamiliar, the quantization guide explains it.
4. Share your result
Once a model runs, your speed helps the next person choose a Mac. Benchmark your own machine and it joins the reports on this site. For the basics that apply to every computer, see how to run an LLM locally.
Frequently asked
How much memory do I need to run an LLM on a Mac?
The model has to fit in the memory macOS lets the GPU use, and that is only part of the total. The table above shows the default share for common sizes. A Mac with more memory runs bigger models, and the calculator checks a specific Mac against a specific model.
Can I raise the GPU memory limit on a Mac?
Yes. The sysctl setting iogpu.wired_limit_mb lifts it, and it resets when you restart. The calculator assumes you leave macOS 8GB when you raise it. On a 16GB Mac that is less than macOS already keeps, so there is nothing to raise.
Is an M-series Mac fast for local LLMs?
It depends on the chip. Speed follows memory bandwidth, and Apple's chips differ widely on it. The best Mac page ranks each chip by the speeds people report.
Which runtime should I use on a Mac?
LM Studio gives you a window and a model browser, and Ollama runs from a terminal. MLX is Apple's own framework for its chips. Each has its own page with the speeds people report on it.