[philistine]

It is getting tiring to hear the same lies over and over again on this site. Extraordinary claims need extraordinary evidence. Show me a clear proof that Apple's performance per watt is reliant to a 200% improvement on the 5nm process. Perhaps I can point you in the direction of the A13 (https://en.wikipedia.org/wiki/Apple_A13), Apple's last 7nm chip. Apple claims their subsequent 5nm node (A14) is 40% more performant than that one.

[smoldesu]

7nm is considerably less dense than 5nm, since it's increased size compounds across two dimensions. That being said, A13 was indeed on 7nm, as well as the A14x they put in the dev kit. I don't think those benchmark numbers would be competitive against 3D Cache Ryzen or current Alder Lake offerings. Furthermore, I question how far Apple will be able to push Apple Silicon. As we've seen with the M2, when Apple can't get a clean node upgrade, they're forced to increase heat/power like Intel and AMD so they can scale their performance. With the upgrade to 3nm being much less dramatic than 10nm -> 5nm (and with Intel catching up rapidly), Apple is under extreme pressure to engineer themselves out of this situation. It may not be technically feasible.

[philistine]

You have in no shape or form supported your claim that Apple gained 200% in performance per watt due solely to the 5nm process.

[smoldesu]

Here, then:

Developer transition kit, running A12z, 7nm:

https://browser.geekbench.com/v5/cpu/search?q=eperm-d995af6e...

Last-gen Mac Mini, running i7-8700B, 14nm:

https://browser.geekbench.com/macs/mac-mini-late-2018

So, let's do some quick math. Assuming the A12z runs in the same power envelope as the M1, it uses roughly half the power of the preceding Intel chip. That makes sense, since both the performance and efficiency cores are using a much denser process. Generously speaking, both of these chips have a similar performance profile, which leads me to believe that Apple's real 'boon' here was simply switching to a denser node. Comparing IPC across ISAs is largely guesswork, but I'm not convinced that AMD or Intel are incapable of manufacturing 5nm chips that decimate Apple's offerings. Apple hasn't allowed it to happen, so we can't say for sure.

[philistine]

Are you comparing an x64 chip with an ARM one and then deciding the only difference is the node process?

We’re talking here about what is unequivocally the most complex manufactured goods in the world. You can’t just boil it down to one thing. Apple made so many decisions that lead them to their extraordinary performance, including financial ones. They heavily paid TSMC for priority access to their new node.

And once again, the gains are nowhere near 200%, your initial claim that is patently wrong and that you refuse to acknowledge as a mistake on your part.

[smoldesu]

I'd love to compare it with a 5nm x86 chip, but thanks to Apple we can only make conjecture. You can agree with me or say my approximations are wrong, but I acknowledged that there's an architectural difference (which is why I used Geekbench instead of comparing useless numbers like clock speed or memory bandwidth). x86's performance on 14nm seems to correlate with ARM's performance on 7nm. Again, if you think that's a BS conclusion; fine. I'm painting funny-numbers with Geekbench scores and approximate power scaling. You'll just have to be contented with waiting until we have a 5nm x86 chip to compare it with, because apparently we can't make ballpark estimates on HN without getting pitchforked.

Apple's chips are fast because they have lots of money. If you disagree with that, then you're no different than the kid at Disney Land who thinks Splash Mountain got built because some genius invented it. We can fight about exact figures until the sun sets, but I think my approximation was fair.