tenthousandthings

M2 spec bump for the current iPad Air before the holidays is one of the few things that wouldn’t be surprising. 

dutchlord said:

Waiting for a M3 XL iMac. And a better ATV with a fresher  GUI and more innovative apps. 

iMac Plus

A spec bump (M1 to M2) for the iPad Air isn't a "new iPad model" -- so this doesn't contradict the likelihood that iMac and iPad Air will both get the M2, via press release, in October.

The iMac jumping directly to M3 never made sense, and contradicted Apple's public statements about every product getting every generation of M silicon.

Gurman pretty much single-handedly engineered the "M3 in October" expectations. I've gotten John Gruber mixed up with Mark Gurman in the past. I think Gruber would be horrified! 

9secondkox2 said:

One thing that puzzles me…

tsmc has been producing 90,000 - 100,000 wafers per month for a while now, so that’s going to be around 50,000,000 or more 3nm SOCs per month. 

And somehow that’s not enough? Even if you factor in let’s say a 50% failure rate, that’s still 25 million SOCs every month. This capacity has been increasing since this time a year ago and the phones have been being assembled for a year already with an estimate of 85-90 million iPhones shipped. That’s a ton of SOCs. Way more than a hit iPhone year plus iPad run Would need. 

And if it’s much less than 50% failure rate since things have improved, that’s quite a surplus of capacity - even with 10% of each wafer going to competitors. 

So where is the constraint? Or is it due to Apple switching from N3b tech to N3E? But isn’t that supposed to increase yield further? 

I don’t think there is a constraint for Apple. It’s a new facility, this “Fab 18” that opened less than a year ago. There are something like a thousand steps in the process(es). TSMC published two papers recently that explain the differences between N3 (N3B) and N3E, and at least one major change was dropped (i.e., moved to N2) in the adjustment.

My guess is that N3 (N3B) was too Apple-specific, and so not flexible enough for other customers. Thus, N3E was born, postponing at least one of the ambitious changes that Apple used for A17. N3 is in high-volume production, but only for Apple. 

If I had to bet, I’d bet on M3 also being on N3. I think the rumor that they are switching to N3E is exactly that, a rumor. The two lines will merge in N2. The mysterious N3S may be a limited, second generation N3B (also Apple-only), used for A18. A19 and M4 will be on N2.

Well put. Quality bloviation, thank you.

While it would still need to be accompanied by the commitments and efforts you mention, there is a way to answer the question posed by Ternus. He’s right that there’s no optimal way to integrate AMD’s current PCIe GPU architecture, for example. Plus, I don’t think he only means unified memory. It’s also about TBDR (tile-based deferred rendering) and Apple’s approach to GPU architecture.

https://developer.apple.com/documentation/metal/tailor_your_apps_for_apple_gpus_and_tile-based_deferred_rendering

Nonetheless, it could still be done, by creating PCIe components that use this approach. Imagination Technologies is one company that could do this, not to mention Apple itself. PCIe 5 and Thunderbolt 5 make this practical.