2020 Mac Mini – Putting Apple Silicon M1 To The Test

Re:2020 Mac Mini – Putting Apple Silicon M1 To The Test by kllrnohj on Tuesday November 17, 2020 @02:59PM (#25128881)

> The 4800u on 5nm is only going to be marginally better.

This is definitely not true. We already know Zen 3 is +20% IPC over Zen 2 on the same process at the same power. So add 20% to the 4800U without changing anything else as a starting point.

Then toss in the process improvements from 5nm (which TSMC says is either 15% faster or 30% less power) as well as any further architectural improvements that AMD is doing in Zen 4 and there's going to be a very significant gap between the 4800U and AMD's 5nm 6800U or whatever they end up calling it.

[ Parent | on HN ]

Re:2020 Mac Mini – Putting Apple Silicon M1 To The Test by bfgoodrich on Tuesday November 17, 2020 @03:02PM (#25128932)

To be clear, I said that a 4800u @ 5nm (if one could simply scale a design like that) would only be marginally better. That the 5nm boogeyman is more incremental than the big advantage it is held as.

You replied that if you take the 4800u, switched it to 5nm, switch it to Zen 3...no actually switch it to to Zen 4 and a completely different chip, it would be lots better so what I said is "definitely not true".

I'm not sure this logic follows.

[ Parent | on HN ]

Re:2020 Mac Mini – Putting Apple Silicon M1 To The Test by kllrnohj on Tuesday November 17, 2020 @03:16PM (#25129168)

A 4800U on 5nm would be 30% more efficient or 15% faster. 5nm was a significant bump.

And that's before considering the density improvement that came along with it (which is also substantial - TSMC's N5 is up to 1.8x the density of N7). Which is why I mentioned Zen 3 & Zen 4, because you don't make the same chip across a shrink. You use the extra budget to do things

[ Parent | on HN ]

Re:2020 Mac Mini – Putting Apple Silicon M1 To The Test by bfgoodrich on Tuesday November 17, 2020 @07:05PM (#25131573)

When a die shrink is correlated with higher performance, generally that means higher clock speed for a given heat profile. The clock speed of the 4800U is already 50% higher than the M1 when running a single-thread task. And despite that, it is 25-50% slower. And it's significantly less energy efficient.

So put that to 5nm. Either you've partly closed the large efficiency gap, or you've closed the significant performance gap, but in neither case will you come close to closing it entirely. So you're either a little less slower, but still a lot less efficient, or a lot slower, but a little less efficient.

Yeah, maybe they'll do some amazing things on the core that'll overcome all of this. But right now the Firestorm core has a massive advantage. AMD can go to Zen4 and 5nm and maybe they'll close the gap, but Apple won't be sitting still.

Apple entered the desktop space and brought the most efficient core (by _far_), hilariously offering the best per core performance without going to magnitudes higher thermal profiles. It's pretty amazing. So now we're into comparing it to hypothetic, mythical alternatives from competitors in the future.

Sidenote: Apple has had blazing cores for a few years now, and every denier will claim it's just some big cache or some other absurd simplification. If that were the case, everyone would just copy them.

[ Parent | on HN ]