Seems the CPU cluster saturates at about 240 GB/s and can't utilize the full memory bandwidth. This bodes well for future clusters with double the number of CPU cores at a node shrink (M2 Max?) or for a Mac Pro (Mac Quadra?).
Maybe. This seems like a cluster-wide limitation - the individual CPU cores can utilize enough memory bandwidth that together they should be able to saturate the bus, but there's some kind of bottleneck on the entire CPU section of the SoCs and who knows how easy or difficult it would be to alleviate that.
Sure, keep in mind that most competing laptops max at 70GB/sec (the never to exceed number), and most desktops are slower than that with 2 channels of DDR4-3200 to 4200 (41-67GB/sec).
So while it's "only" 240, that's an excellent number. Keep in mind that you generally never see 100% of theoretical bandwidth.
