Which is why I suggested comparing to the EPYC 73F3, which is a 5950X clocked at 3.5 - 4 GHz, with 4x the L3$, 4x the memory bandwidth (if you don't overclock i

Which is why I suggested comparing to the EPYC 73F3, which is a 5950X clocked at 3.5 - 4 GHz, with 4x the L3$, 4x the memory bandwidth (if you don't overclock it), and 5~6x the IO bandwidth. We know a 5950X is roughly on-par with an M1 Max (at least ignoring the latter's 2 efficiency cores). If the occasional wins of the M1 Max are due to memory bandwidth, this should more-or-less turn the tables.