Is memory bandwidth partitioned in any sense? As in, is it possible that the bandwidth to a core, or a socket, or some grouping of CPUs, is not the total memory bandwidth?

@anon33 I would guess not because from my understanding, memory operations from all the different cores still have to share the same memory bus that links the chip to main memory.

I wonder if there's any advantage to having separate memory buses to different memory banks and explicitly linking to different CPUs. Put another way, is there any slowdown to having larger memory and instead separating it out into smaller segments which we can potentially search fetch from faster? Or is this largely negligible compared to other costs.