Unified Memory Architecture (UMA)
Shared physical memory pool for CPU, GPU, and Neural Engine on Apple Silicon
Apple
In plain English
One giant memory swimming pool where the CPU and GPU swim together without having to haul buckets of data across a bridge.
Technically
A shared physical LPDDR5/LPDDR5X memory pool accessible by CPU cores, GPU execution units, and Apple Neural Engine (ANE) with up to 800 GB/s bandwidth on M-series Max/Ultra chips. Eliminates host-to-device PCIe transfer overhead.