can you experts stop those pointless arguments about mid vs high tier for just 1 min and explain the numbers to us non “pro” folks. it doesn’t make sense the max performance is near identical to pro….isn’t max suppose to be 2x the performance of pro in multi core and gpu benchmarks? Thanks
Since all M-chips are assembled based off the same “units” (CPU, GPU cores, neural engines, encoding engines, RAM lines, etc) it might happen that two chips might have same scores on a specific “chip block”:
- All M2s use the same single CPU cores that make up their 8(M2) / 10(M2Pro) / 12 (M2Pro) / 24 (M2Ultra) variants: single core scores will tend to be identical. Even the iPad Pro’s M2.
- If a single core score is considerably higher (say 5%+) there’s maybe some clocking trickery going on for those cases (like Intel’s turbo boost, basically dynamic overclocking on the fly). Plus maybe cache benefits between models. However, for example, L1 and L2 cache levels of M2 Pro and M2 Max are identical, except the L3 where M2 Max is doubled.
- The full M2 Pro 12 core and M2 Max 12 core are basically assembled the same on that CPU front. Synthetic geekbench benchmarks aren’t likely to show any difference there (blind to memory bandwidth and minor cache differences). Same scores.
- But you are definitely right on those GPU scores. Up to 19-cores on M2 Pro and up to 38-cores on M2 Max should yield double the performance. And it tends to, render times on 3D applications are effectively close to halved (when properly developed) or games tend to run a lot better (if GPU was indeed the bottleneck). Those geekbench scores are definitely missing something.
- The rest of the difference is the M2 Max having more of the extra hardware modules and controllers: more encoding/decoding video engines, more memory bandwidth, more thunderbolt controllers, more support for hardware monitors, etc which won’t show at all on the benchmarks.
I’m no expert on any of this in any case, this is what I have come to understand to date watching videos and having two of them.
Sometimes in some applications the performance far from doubles, as was the case for M1 Ultra (supposedly 2 M1 Max). Rendering on Blender Cycles raytracer shows barely 30% extra performance between M1 Max and M1 Ultra. M2 Max is even close to the same as M1 Ultra on that front!
Supposedly the M2 architecture got revised for fixing that “scaling issue”… we will see when the M2 Ultra real life use cases come to light.
Hope that helps to make sense of those scores.
Good for a generic baseline, far from accurate for key real life scenarios… check some of the more complex anandtech benchmarks if time permits, I don’t really know what the score numbers mean in a lot of them but it’s a lot more thorough and the M1s and M2s at the time had numbers in
some of the categories that would compete or obliterate the most expensive or most massive workstation CPUs (some could cost more than a single Mac Studio Ultra, just for the CPU).