WWDC wraps up
And a new 'S' box from Xbox
Welcome back to Multicore for Tuesday, June 13th.
I'm afraid we aren’t quite done with Apple just yet. The fallout from last week’s Worldwide Developers Conference continues to, well, fall out.
Every year, Daring Fireball's John Gruber does a live episode of his podcast The Talk Show at WWDC where various Apple executives chat about the week's announcements. This year’s show featured hardware SVP John Ternus, software SVP Craig Federighi, marketing SVP Greg Joswiak, and Mike Rockwell, who runs Apple’s AR and VR group and led development on the Vision Pro.
The vibe is more fireside chat than hard-hitting interview, but I did appreciate how Gruber pressed on much of what I wrote about in my issue on the Mac Pro last week.
Gruber: Obviously industry wide, the theme of the year clearly is AI and AI training. And that whole area, it seems like all of the compute takes place on graphic cards. And so I'm just wondering with this new Mac Pro, what is the message to Mac users, pro Mac users who are looking to experiment, or if it's actual work to do their own AI training. Is the idea that they're best off renting GPUs in the cloud and doing that, that that's not something you envision as something you do on the desktop in front of you? Is the answer that they should buy a Windows PC?
Ternus: That’s never the answer. *laughs*
Joswiak: Look, Nvidia is doing a good job with that. There's no doubt about it. We wanted to focus on the things that are most important to our customers, and we do those things well. And the things we can do with our unified architecture and with our M2 Ultra are things that nobody else can do and no other personal computer can do. And so we have our strengths, they have theirs — they're doing a good job, you know, great for them. But we've got stuff that no one else can do.
Gruber: Are there technical barriers to having expandable graphics through PCI that would be only used for compute as opposed to video? Or is that just a design choice?
Ternus: I mean, fundamentally, we've built our architecture around this shared memory model and that optimisation. And so it's not entirely clear to me how you'd bring in another GPU and do so in a way that is optimised for our systems. It just hasn't been a direction that we wanted to pursue.
The context to this exchange is that while the new Mac Pro does have PCIe slots, you can’t use them for GPUs or memory. The M2 Ultra chip, which is also found in the much smaller (and cheaper) Mac Studio, is all you get for GPU compute, and the RAM is soldered onto the die.
Joswiak and Ternus’ answers are reasonable for what they are — and I did not expect an Apple exec ever to give unprompted props to Nvidia — but they’re really not a answer for Mac users who want to work on AI. Nor is this an answer for anyone who was using well beyond the new Mac Pro’s maximum RAM capacity of 192GB — the Intel model could go up to a terabyte and a half.
Gruber went on to ask about whether Apple is still interested in pursuing speed and making “the race car of computing”:
Keep reading with a 7-day free trial
Subscribe to Multicore to keep reading this post and get 7 days of free access to the full post archives.