Heikki Kultala (heikk.i.kultal.a.delete@this.gmail.com) on June 2, 2022 10:45 am wrote:
> Anon (no.delete@this.spam.com) on June 2, 2022 12:35 am wrote:
> > Eric Fink (eric.delete@this.anon.com) on June 2, 2022 12:20 am wrote:
> > > I would agree with everything you said, but there is still the fact that M1 — while prioritising
> > > power consumption — is essentially reaching identical performance as the x86 chips. Sure, binned
> > > desktop Alder Lake is marginally faster (~20% , but it also pays a significant ~5x penalty in power
> > > consumption to get at most ~20% higher ST performance. So I am not sure about the accuracy of the statement
> > > that Apple prioritises power over performance — kind of seems to me that Apple gets both.
> >
> > Apple is using TSMC 5nm while Intel is using 10nm which they call 7nm
>
> Totally wrong.
>
> Apple is using TSMC ~15nm process which TSMC calls "N5". N5 is just a name.
>
> Intel is using ~18nm process which they originally marketed as "10nm"
> and now call it "Intel 7". Those 10 and 7 are only names.
>
> > and AMD uses TSMC's 7nm, and both Intel Intel and AMD supports SMT.
>
> Wrong. AMD uses TSMC's 20nm process which TSMC calls "N7". THis N7 is only a name.
>
> And SMT has NOTHING to do with single-thread performance.
>
> There is no 7 nm or 5nm anywhere.
>


Where are you getting these claims about "~18nm" and "20nm" processes? Those claims are no more real than "5nm" or "7nm".

The real figure of merit is transistor density, and while different companies might use different methods to calculate their claimed densities, it is the best data we have (actual density in shipping products is useless because higher power transistors are lower density, cache scales differently than logic and so on so comparing e.g. M1 vs Adler Lake based on claimed transistor count (and even that has different methods to compute) divided by die size doesn't get you very useful information.

So we can consider "Intel 7" (f.k.a. "10nm") and TSMC's "N7" to be similar. I don't know Intel's claimed density for Intel 4 but I'd guess it is roughly comparable to TSMC's N5 or N4.

Too many people want to get hung up on "but it isn't REALLY 5nm" and ignore what matters. TSMC publishes information on performance improvement (when holding power constant) and power improvement (when holding frequency constant) so it is pretty easy to see that the gains from N7 to N5 are not that great - certainly nothing remotely close to the 70-80% gain in transistor count. Unfortunately Intel and Samsung don't make such detail as readily available, but maybe that will change for Intel now that they are finally starting to get serious about becoming a foundry.