it probably depends on particular software and how well its optimized for AS. octane might be 3x faster on m2 ultra over m1 ultra because it takes really good advantage of features in m2, or because it had severe limitations with m1 and had previously been leaving performance on the table.
but yeah i agree its probably gonna get around mid 3000’s on the blender bench. interestingly, while thats still well behind nvidia, the highest benching result i can see for amd is around ~3600? so two generations in, they’re approaching amd’s best levels of performance (save for multi gpu systems).
i don’t think they’re ever going to try to surpass nvidia. i’m guessing their goal is what they’d consider a balanced system that works for “most” user needs. but if they’re serious about spatial computing, i hope that involves giving as much love to 3d performance as they do to video acceleration these days.