For Redshift3D, scaling is actually worse. My m1 max does the bench in 10 min flat. A full Ultra about 7m 30s. To be honest, I (and others) hoped that the Ultra would have scaled better than linear in RS due to increased bandwidth etc so it was quite sad to see the numbers trickle in. Octane and Cycles seems to be best in class right now and giving about the same perf as a single 6800 (not XT!) for the Ultra. I guess we have been through this before in this long thread but extrapolating from rumours etc gives that we need something more than just an evolution of the Apple silicon series in the GPU department. Heck, all other gpu, even Intels Arc, have RT cores now. Macs have gone from having GPUS (from AMD) with ballpark specs of the windows competition (or maybe 50% at worst?) to being at least 2 full generations behind. The CPU is really nice so it is sad that it cannot be matched with an equally impressive GPU. Let's assume we get RT cores on the Max and Ultra chips and they solve the scaling to be linear or at least 90%. Best case improvement for the M2 Max would then be 40% increase due to 4x10 cores with slightly higher clocks giving us about the same perf as a current 48 GPU core m1 ultra. Doubling that (requires good scaling!) for the m2 ultra we are at about 2500 point in the linked blender bench. Or about the same perf as as nvidia 3060. If Apple can add RT cores for a 2x perf boost and some "extra fixes and optimisations " the Mac might be relevant again considering power draw. Let's hope.
devtalk.blender.org
