I would assume a lot of it comes down to power budget per core like the 64 core, but holy cow that total wattage. “Monster” results, indeed.
I’m actually curious how much - or little - benchmarks would be affected by actually running multiple instances simultaneously, because that’s kinda the whole gig.