programmer

DoubleJac said:

in the report that claimed Apple bought all of TSMC‘s 3 nm, there’s nothing about it all being for the A17 Pro. that includes the link in the article.

The animation of the Apple and the Mac Finder face logos heavily imply hardware ray tracing coming to the Mac. Also remember, the M1 Pro was “scary fast” and the M1 Max was “scary faster” in 2021.

I was going to say the same thing about 3nm capacity.  Whether the logo says anything about anything other than Apple’s art department, we shall see.  And most of their chip announcements have been fancy pre-recorded videos anyhow, so another shouldn’t be a surprise.  

The A17 includes the new GPU w/ hardware raytracing capabilities.  That will be in the M3 series of chips.

I think the odds of a significant GPU revision are more likely in M3 than you think, primarily because it was initially aimed to happen in the M2 generation but wasn't ready in time.  They've had quite a long time to work on it, and there is a lot of market pressure to get it into product.

I'm a bit more optimistic about the M3 timeline as well.  Earlier in 2024, perhaps.  End of 2023 seems too optimistic... although we do know that Apple has booked to use all of TMSC's 3nm capacity in 2023.  So either their capacity isn't too much, or Apple is expecting to sell a lot of new A17s (but the smartphone market has been softer the past few years), or it will be a combination of A17s and M3s.

saarek said:

You’re right about performance, the Threadripper will no doubt be faster, but it will also consume far more power for said performance.

I wonder if Apple is cooking up a method of multiple ultra class SoC in the same system. Imagine 5 daisy chained M3 Ultras (or even better an unannounced Ultra Max) in the Mac Pro.

Yes, I think this is really their only choice, if they want to compete in that part of the market.  

They might be able to push clock rates a bit higher, but their chips are primarily tuned for low power mobile applications and retuning everything to compete with the AMD and Intel (and nVidia) blast furnaces doesn't make sense for a machine that is a niche product for Apple.  Plus their whole story is about efficiency.  So how to make a more powerful machine without a huge amount of effort?  Make it modular, put their chips in modules (each module being a full-blown CPU/GPU/Memory/IO unit), and pack them in together.  With lower heat output and power requirements, they can do this better than anyone else.  

It becomes a software problem then -- how to make use of so many discrete processors in a single box?  If they know what their target markets are (e.g. ML/AI, render farms, Xcode cloud, etc) then they can write software (applications, libraries, drivers) that addresses each of those.  And if they have software that makes use of what are essentially "headless servers" then selling these units into clouds or server farms suddenly becomes sensible, which is a market Apple hasn't had access to for a long time.  

The real question is whether Apple is actually interested in going after these markets.  That's anybody's guess.  They've been willing to walk away from the very high end of the market before, and haven't done servers in a long time.

Well this fits with Apple pausing M-series chip production.