The main ways that I think the architecture of M1 is going to affect memory usage are probably:
Caches and superscalar execution do not reduce RAM pressure. Data in the CPU's caches, and instructions in the execution pipeline, are copies of the data in main RAM. Data still takes up space in main RAM, even if it's in the cache or currently being executed.
I will be very surprised if a thorough analysis finds a significant real difference in memory usage between ARM and Intel. I could be wrong, though!
- ARM has lower instruction density, so ARM code is going to take up a small amount more RAM.
- Unified memory might allow merging of copies of GPU buffers - but other integrated GPUs also use system RAM.
Caches and superscalar execution do not reduce RAM pressure. Data in the CPU's caches, and instructions in the execution pipeline, are copies of the data in main RAM. Data still takes up space in main RAM, even if it's in the cache or currently being executed.
I will be very surprised if a thorough analysis finds a significant real difference in memory usage between ARM and Intel. I could be wrong, though!