Unified memory
A single pool of memory shared by the processor and graphics, as on Apple Silicon and some mini-PCs, so a large share of system memory can hold a model. It lets machines with no discrete GPU run models that would otherwise need one.
On a conventional PC, the GPU has its own VRAM, separate from system RAM, and a model has to fit in that VRAM. Unified memory, used by Apple Silicon and some mini-PCs such as the Ryzen AI Max, instead gives the processor and graphics one shared pool, most of which can be used to hold a model.
This is why a Mac with 64GB or 128GB of unified memory can run large models that no single consumer graphics card could, though usually at lower generation speed than a fast discrete GPU. The trade-off is memory capacity and low power against raw throughput. See GPUs and Apple Silicon.