Marco Arment

You did just gloss over those two numbers pretty quickly.

You did just gloss over those two numbers pretty quickly.

Not to mention the process of replacing it would probably not be simple. Right, exactly.

Not to mention the process of replacing it would probably not be simple. Right, exactly.

Yep. This is a profit center. I mean, look, this is like when you buy parts for an expensive car from the manufacturer. If you're going to have some kind of high-end sports car, if you want to go to the manufacturer to buy their brake pads, they're going to cost a heck of a lot more than everyone else's brake pads, but that's kind of their privilege.

Yep. This is a profit center. I mean, look, this is like when you buy parts for an expensive car from the manufacturer. If you're going to have some kind of high-end sports car, if you want to go to the manufacturer to buy their brake pads, they're going to cost a heck of a lot more than everyone else's brake pads, but that's kind of their privilege.

Some of it is like, well, we have very high specs for our brake pads, but a lot of it is just their profit, right? That's the business model, and that's what this is, too.

Some of it is like, well, we have very high specs for our brake pads, but a lot of it is just their profit, right? That's the business model, and that's what this is, too.

I mean, they're also, in all fairness to them, this might not be purely economic. This might also be, you know, maybe they made these prototypes and they weren't as good as they thought. Because if you look at the Ultra chip, which is the two Maxes, you know, talking to each other over the silicon interposer. It does not scale perfectly in all useful ways. There is some degree of overhead.

I mean, they're also, in all fairness to them, this might not be purely economic. This might also be, you know, maybe they made these prototypes and they weren't as good as they thought. Because if you look at the Ultra chip, which is the two Maxes, you know, talking to each other over the silicon interposer. It does not scale perfectly in all useful ways. There is some degree of overhead.

The scaling is not complete. You don't get 2x of everything when you do that. You get close to 2x in certain ways, but sometimes more like 1.5x in other ways. So obviously, they're not actually achieving perfect scaling. There's other bottlenecks somewhere or the limitations.

The scaling is not complete. You don't get 2x of everything when you do that. You get close to 2x in certain ways, but sometimes more like 1.5x in other ways. So obviously, they're not actually achieving perfect scaling. There's other bottlenecks somewhere or the limitations.

maybe when they tried a 4x design it just wasn't worth it because maybe a 4x design rather than reaching 4x performance maybe it only reached 2.5x performance or something like that like maybe it just wasn't technically compelling and you know would you pay for four times as much or whatever. I'm sure the cost would be absurdly more expensive.

maybe when they tried a 4x design it just wasn't worth it because maybe a 4x design rather than reaching 4x performance maybe it only reached 2.5x performance or something like that like maybe it just wasn't technically compelling and you know would you pay for four times as much or whatever. I'm sure the cost would be absurdly more expensive.

Would you pay that much for, quote, only two or three times the performance? Maybe not. And maybe they decided the better way to go was maybe to wait for something like a chiplet design where they could do things like The parts that scale well offer more of those, like the GPU cores maybe, and maybe not duplicate the entire chip design and all the other ways in order to achieve more GPU cores.

Would you pay that much for, quote, only two or three times the performance? Maybe not. And maybe they decided the better way to go was maybe to wait for something like a chiplet design where they could do things like The parts that scale well offer more of those, like the GPU cores maybe, and maybe not duplicate the entire chip design and all the other ways in order to achieve more GPU cores.

When you look at CPU performance, the CPU performance I think is probably what scales the worst as you add more cores here because of various limits and stuff. What the Mac Pro needs to be more competitive with its previous market as well as the current market, what it needs is more neural engine cores and or more GPU cores for certain configurations.

When you look at CPU performance, the CPU performance I think is probably what scales the worst as you add more cores here because of various limits and stuff. What the Mac Pro needs to be more competitive with its previous market as well as the current market, what it needs is more neural engine cores and or more GPU cores for certain configurations.

So if they're moving to a system of chiplets, which sounds very promising...

So if they're moving to a system of chiplets, which sounds very promising...