A big part of the reason is that the CPU, GPU, and neural engine all have direct access to the memory. That's the idea behind the unified memory that Apple talk

A big part of the reason is that the CPU, GPU, and neural engine all have direct access to the memory. That's the idea behind the unified memory that Apple talked about during the keynote. Compare this to a normal setup, where you have main memory and dedicated VRAM. You load something into a game by fetching it into RAM, and then loading it into VRAM by dispatching it over PCIe to the graphics card processor, which then stores it in VRAM. There's a lot of extra steps. My guess is that their memory subsystem is fairly radical to handle the different types of memory access for the different functional subsystems, so they optimized for specific memory modules. I really would love to see some more documentation on the M1 memory subsystem, there are so many unknowns behind the very impressive performance numbers we've seen over the past few days. And I personally suspect that the future will have traditional ram, with the on-chip ram acting as a kind of last level cache. You can't really scale this to 64GB or higher without some major thermal problems.