Comment by hedora
1 day ago
The article says it’s going on Intel 18a, which has better performance, but worse density/power consumption than TSMC’s comparable node:
https://www.techspot.com/news/106782-intel-18a-found-faster-...
Unless Apple is going to add active cooling or something to the Air, iPad and iPhone, I’d expect more thermal throttling on the intel chips (though the difference isn’t as stark as I assumed — it looks like Intel closed the gap a bit with 18a).
The performance / power difference between Intel 18a and TSMC 3nm / 2nm is not going to be that wide. Certainly not enough to require active cooling on phones / tablets. The rumor says they are targeting the M series processors anyway, so probably looking at it for their laptops first where there is active cooling and the power envelope is more forgiving.
TSMC's 2nm is much, much higher demand. Apple's probably getting priced-out of the node by Nvidia in much the same way Apple bought out TSMC 5nm before.
Apple's cheaper products and hardware cash cows can't afford to pay that sort of tax, so it makes sense to boot them onto 18A. Those are the binned products that would have been throttling anyways.
TSMC and Apple have a long standing partnership, and I doubt TSMC is dumb enough to throw a massive amount of long term business away in exchange for short term AI bubble gains.
TSMC has even been cutting Apple special terms they don't give others:
> every time TSMC introduces a major upgrade, called an advanced process node, to its chipmaking, the defect rates of the dies stay relatively high until it can iron out the kinks. For 3 nm, the most cutting-edge node launching this year, the yield on wafers has recently been in the range of 70% to 80%, according to analysts, as well as one person with direct knowledge of the process.
That number would be a tough pill to swallow for TSMC’s customers, which typically pay for the wafer and all of the dies on it—including the bad ones. But in a break from standard practice, the Taiwanese manufacturer has only been charging Apple for dies that work—“known good dies,” in industry parlance—these people said.
https://archive.ph/yfGLp
I doubt TSMC has a choice. Any fab allocation Apple can pay for, Nvidia can afford twice over. With that kind of money Nvidia could swallow the cost of 40% broken dies and still turn out higher hardware margins than any 2nm iPhone ever could. Neither the iPhone Air nor Vision Pro justify Apple's push to dominate the latest nodes.
Novelty applications like "performance smartphone hardware" will have to wait on the sidelines. The datacenter needs it more than Apple or Qualcomm, and they've brought the beaucoup bucks to prove it.
9 replies →