Rumor: “M2 Pro” on 3nm?

[Cmaier]

Report: TSMC to start mass production of new 3nm 'M2 Pro' chip later this year — 9to5Mac

Apple this week unveiled the new M2 chip, which is the first upgrade to the Apple Silicon chips created specifically for the Mac and iPad. While the new M2-powered MacBooks are yet to hit the stores, analyst Jeff Pu of Haitong Intl Tech Research has now reported that Apple supplier TSMC will...

apple.news

I suppose it’s possible, since the effort to port to 3nm would be used to create the Pro, Max, Ultra, Ultra Max, etc. It would be an unusual maneuver for Apple, of course. (And a lot of the effort they made to create M2 on 5nm would not be maximally-leveraged). That said, I‘ve worked on chips where we ported midstream like this, so anything is possible.

However, quoting the article:

For those unfamiliar, the scale of a chip in nanometer represents the distance between its transistors. The smaller the distance, the better the performance.

I don’t agree with any of that

 

[theorist9]

In April I read an article by Anton Shilov at Anandtech saying TSMC will start production of its N3 chips the 2nd half of 2022, and deliver those to customers (which means Apple and possibly Intel, since they've reserved the N3 production) in early 2023 (see https://www.anandtech.com/show/1735...n3e-in-2024-n2-in-2026-major-changes-incoming). I'm sure you've read the same article, and others. So we already expected to see N3-based Pro/Max products in 2023. Unless I'm missing something, Pu seems to be saying the same thing, except he's calling these N3-based Pro/Max chips "M2" instead of "M3". I.e., I don't think Pu's analysis speaks to whether or not we'll be seeing 5 nm N4P Pro/Max products at the end of this year, he's just saying we'll be seeing N3 Pro/Max products (which we expect to arrive 2023), and (likely incorrectly) calling those M2.


From Shilov's article:
"TSMC is set to start ramping up production of chips using its N3 node in the second half of the year and will deliver the first commercial batch to a client (or clients) in early 2023, which is when it will receive the first N3 revenue."

 

[B01L]

M2 (5nm+) might just be all the low-end M2 SoCs, and late 2022/early 2023 might bring us M3 (3nm) and the higher end SoCs (Mn Pro/Max/Ultra/Extreme)...?

 

[Cmaier]

Yeah my comment was only that it seems weird to have M2 variants (meaning avalanche blizzard cores) on 3nm. Not impossible. Not even all that unlikely. Just not something that apple has done before unless I’m forgetting some prior example.

 

[Andropov]

'Later this year' almost necessarily means 'after A16 devices have been released'. Could it be using the A16 cores? If the A16 is built on 3nm too, why not use those cores then?

 

[B01L]

So what we want is A16-based M3 on 3nm with hardware raytracing, in first half of 2023...?

 

[Cmaier]

'Later this year' almost necessarily means 'after A16 devices have been released'. Could it be using the A16 cores? If the A16 is built on 3nm too, why not use those cores then?

That would make more sense, but then based on precedent those would be named M3 and not M2. Also, so far, Apple doesn’t seem to mind having iPhone use newer cores than its computers.

Or maybe Apple wants to test the 3nm fab using lower-volume M-series chips using known-good cores, so that they don’t have to deal with two different sources of problems. Or maybe M2 Pro and Max and Ultra are indeed going to be 5nm. Or maybe Apple designed M2 on a lowest-common-denominator grid so that scaling is simply a mathematical transform. (That’s how we did it one time at AMD - I believe it was the original opteron that was designed that way).

Who knows.

 

[Cmaier]

So what we want is A16-based M3 on 3nm with hardware raytracing, in first half of 2023...?

I would love one of those on that timetable - apple would get me back on my old-time habit of replacing my MBP every 2 years or so

 

[Cmaier]

And a guy who has sometimes been right says:

https://www.twitter.com/i/web/status/1535259238183800832/

If he’s suggesting there’s a chance M2 Pro/Max will get v9, I will bet against him on that.

 

[theorist9]

And a guy who has sometimes been right says:

https://www.twitter.com/i/web/status/1535259238183800832/

If he’s suggesting there’s a chance M2 Pro/Max will get v9, I will bet against him on that.

Agreed. Plus I don't understand Shrimp's position. Is he thinking the first N3 chips will be A15-based (like the M2 Air's) rather than A16-based, and thus will be called M2 instead of M3? I thought the N3 chips were supposed to be A16-based (at least).

 

[Cmaier]

Agreed. Plus I don't understand Shrimp's position. Is he thinking the first N3 chips will be A15-based (like the M2 Air's) rather than A16-based, and thus will be called M2 instead of M3? I thought the N3 chips were supposed to be A16-based (at least).

Yep, it’s muddled. It’s always possible that M2 Pro/Max are on 3nm, but if they are, there won’t be a big architectural change.

If Apple *does* go to 3nm somehow on those chips, the performance/watt vs. Intel/AMD would be crazy for a year. But, again, they’ve never done this sort of thing before so I am skeptical.