Correlation between Geekbench 6 MacOS GPU Metal score, and Geekbench AI
So I have a MBP M1 Max, and am looking to see if I can replace it with an M4 Pro, specifically for machine learning compute (PyTorch MPS training & inference).
Browsing the Geekbench 6 GPU Metal scores shows the M4 Pro coming in under the M1 Max, although close (112,726 versus 124,019) which makes sense; because it is 20 GPU cores versus 32 GPU cores, 273GBs versus 400GBs memory bandwidth, even when the M4 is a newer and faster GPU.
But then when I look at the Geekbench AI Core ML GPU scores, the M4 Pro scores are now noticeably higher than the M1 Max scores, specifically 15,127 / 16,407 / 15,159 versus 11,480 / 12,524 / 11,080; so the M4 Pro scores 30% higher!
I appreciate the Geekbench AI scores are not just focussed on raw compute but also look at precision, so are weighted. But these scores suggest that for machine learning workloads, and in spite of the lower memory bandwith and lesser GPU cores, the M4 Pro performs around 30% better!
Clearly I am missing something... thoughts please?
Keyboard shortcuts
Generic
| ? | Show this help |
|---|---|
| ESC | Blurs the current field |
Comment Form
| r | Focus the comment reply box |
|---|---|
| ^ + ↩ | Submit the comment |
You can use Command ⌘ instead of Control ^ on Mac