Unified memory
Also known as: shared memory
Unified memory is Apple’s architecture where the CPU and GPU draw from the same pool of RAM.
The payoff for LLMs: a 64 GB Mac can put most of that 64 GB behind the model, running 30B–70B models that simply won’t fit on a 24 GB consumer GPU.
- Bigger models than a comparable GPU.
- Lower throughput than a top data-center GPU.
The trade in one line: unified memory wins on *model size you can load*; a dedicated data-center GPU wins on *speed per token*.