unified memory
One pool of RAM shared by the CPU and GPU, as on Apple Silicon, instead of separate system RAM and dedicated video RAM. It means a Mac with 128 GB can hand almost all of it to a model, with no copying between chips. macOS caps the GPU-usable share at roughly 65 to 75% of total.
There are two ceilings on a Mac and the lower one wins. The first is the GPU share, roughly 65 to 75% of the pool, so 128 GB of RAM is nearer 90 GB of model. It is a soft limit rather than a physical one and it can be raised. The second is everything else running, because the model and the OS and the browser all draw from the same pool.
A graphics card works the other way around. Its VRAM is the model's alone and nothing competes for it, so the number on the box is the number you get, but crossing it spills the model into system RAM at around 80 GB/s and costs an order of magnitude. Apple Silicon slides down that hill gradually. A card falls off it.