I had a random thought while out for a walk today. The M1 and the A14 reportedly use the same core design for their performance and efficiency cores. The A14 has two performance cores and four efficiency cores. The M1 has four of each. We have multicore Geekbench scores for both the A14 and the M1. So, with a little algebra, we can figure out how many points each type of core contributes to the multicore Geekbench.
2p + 4e = 4203
4p + 4e = 7689
p = 1743
e = 179.25
Now, this comes with a boatload of caveats. Modern processors are complicated and there are reasons even the multicore version of Geekbench might not scale exactly linearly with the number of cores as I’m assuming. Not to mention the fact that I’m completely ignoring the fact that the M1 is clocked slightly faster than the A14. I’m just going for a very rough estimate here.
With those caveats in mind, let’s take a look at the rumored MacBook Pro chip with two efficiency cores and eight performance cores. Plug in the numbers above and you get a theoretical Geekbench multicore score 14302.5.
Just to put that in perspective, here’s the Geekbench multicore Mac leaderboard:
I would expect M1X to be slightly better than this - due to engineering changes. And M2 much better - due to even more engineering changes and the generation-to-generation general speed up.
(Though I do Performance for a living, I think your numbers, the numbers for my-soon-to-arrive iPad Pro, my thoughts above, are all way above what I probably need. But then not everybody needs a 190-way z15 mainframe.)
It depends on if apple want to push into the pro angle. For geek bench likely it will not matter but on the M1 a single firestorm core can consume almost the entire memory bandwidth available to the chip. So the multi-score performance (in tasks that can use it, not geek bench) is bottlenecked in the M1. If for the M1X apple opt to provide enough bandwidth for all cores to run un-bottlenecked then in real world memory heavy tasks the M1X will dominate the entire industry. A single M1 firestorm core can chomp through almost the same amount of data per second as a 16core top fo the line AMD desktop chip!
There is also a chance that apple have a even larger core i like the name Plasma Storm that uses more die area to push single core performance up (you get better perf/W in single core if you increase the die size rather than frequency). They would however likely not fill the chip with these since in multi threaded tasks your likely going to get better perf/W and perf/die area having more smaller firestorm dies.
PS. I expect it might be called P1 or J1 not M1X this way they will have a clear letter for each level of the family S* (watch) A* (phone, small iPad) M* (ultra portalable) P* (pro) this will let them have a different refresh frequency without them looking old.
Increase die size, reduce feature size, use real estate more effectively - all are more likely to produce speed ups than increasing frequency. (And also likely to be more power efficient.)
Again, to mention z15 as an example, we’ve upped engine speed 10-15% per generation on the mainframe within the same power envelope and without increasing the clock frequency. All through processor design. So, I’m pretty sure M1X and M2 and maybe a few more generations of Apple Silicon could do something similar.
)