[Bloomberg] Apple starting process to dump Intel in Macs

[Gideon]

IMHO simply for the mac line sales are too low to justify custom chips unless they greatly reduce the lineup so that 2 chips/socs are enough but that would kill any mac pro type thing. and even then mac sales are tiny compared to iphone but I guess they would save a ton on the software side to make it worth it.

IMO they can just postpone switching the Mac Pro lineup from X86 (or not switch at all for the forseeable future). These are the platforms using the most X86 coupled code (adobe stuff, etc) and the least likely to use the shared apps between iOS and MacOS.

Sure, not switching means that they will have to keep supporting 2 versions of MacOS for a while longer, but if Windows can support Win10 on arm ... why not?

 

[coercitiv]

IMO they can just postpone switching the Mac Pro lineup from X86 (or not switch at all for the forseeable future). These are the platforms using the most X86 coupled code (adobe stuff, etc) and the least likely to use the shared apps between iOS and MacOS.

Macbook Pro users are in the same boat though - lots of professionals doing design, coding, and video editing on some of the more powerful laptops.

The 12" Macbook and the Macbook Air may be the perfect candidates for this transition, but the 15" Macbook Pro is a different breed, with people running specialized software on them. Many use them as desktop replacements too.

 

[jpiniero]

Sure, not switching means that they will have to keep supporting 2 versions of MacOS for a while longer, but if Windows can support Win10 on arm ... why not?

You'd be surprised how much is shared between iOS and OSX as it is.

They could keep the Mac Pro on x86 and just use Epyc instead I guess for the time being. Apple is doing a lot around keeping iOS App Store apps portable so perhaps with Marzipan they will extend it to OSX as well... so that as long as you use the Mac App Store for all your apps Apple will take care of the portability issues.

 

[ksec]

And Thunderbolt.... is still not opened up.

 

[JDG1980]

I think there's a very real chance that Apple will surprise everyone with a Transmeta/Loongson-type chip that can not only execute existing ARM code, but x86 code as well. The die space requirements would not be all that huge because the x86 front end decoder is a tiny part of a modern CPU; the real work is done on the back end in micro-ops, and this can be shared between both architectures.

What about Intel's patents? Well, most recent opcodes fall in the "nice to use, but not essential" category. The last x86 extension that is really a must-have is SSE2 (you can't run current versions of Windows without it, for example). SSE2 debuted on the Pentium 4 in 2001, so the patents should be expiring around 2021, maybe a bit sooner or later depending on when they were actually filed. (The law since 1995 has been 20 years from filing.) So after that happens, Apple (or anyone else) can simulate, emulate, or implement in hardware all x86 features up to and including SSE2 without having to make any license agreements. They would have to license AMD64 (debuted in 2003) if they wanted 64-bit support, but AMD can license this on their own since they invented it, and doing so would basically be free money for AMD since Apple buys their CPUs from Intel currently.

 

[moinmoin]

And Thunderbolt.... is still not opened up.

USB4 will be USB 3.2 + Thunderbolt 3.
https://www.anandtech.com/show/14048/usb4-specification-40-gbps-type-c-tb3

 

[darkswordsman17]

I think there's a very real chance that Apple will surprise everyone with a Transmeta/Loongson-type chip that can not only execute existing ARM code, but x86 code as well. The die space requirements would not be all that huge because the x86 front end decoder is a tiny part of a modern CPU; the real work is done on the back end in micro-ops, and this can be shared between both architectures.

What about Intel's patents? Well, most recent opcodes fall in the "nice to use, but not essential" category. The last x86 extension that is really a must-have is SSE2 (you can't run current versions of Windows without it, for example). SSE2 debuted on the Pentium 4 in 2001, so the patents should be expiring around 2021, maybe a bit sooner or later depending on when they were actually filed. (The law since 1995 has been 20 years from filing.) So after that happens, Apple (or anyone else) can simulate, emulate, or implement in hardware all x86 features up to and including SSE2 without having to make any license agreements. They would have to license AMD64 (debuted in 2003) if they wanted 64-bit support, but AMD can license this on their own since they invented it, and doing so would basically be free money for AMD since Apple buys their CPUs from Intel currently.

I'm not sure it'll even be that complex. And I think we've already seen precedent for what Apple is likely to do when they switched from PowerPC to Intel x86. They'll just do software translation while they optimize the code for the new chips as they update the software over time. They'll rely on the inherent efficiency advantage (and in portable markets, its likely they'd have an outright performance advantage over the Intel chips they've been using, so they'll have some performance to spare so if some is lost due to translating code, not a huge issue and their efficiency advantages means it won't kill battery life), and then rely on brute forcing things until it gets optimized, wherein it'll likely see big performance improvements (that Apple can tout as a big upsell, much like how they touted that stuff back when they transitioned from Power to Intel, and have been touting in iPhones/iPads).

I think we see them start with the iPad Pro, Macbook, and Macbook Air lines. These are relatively small niches, but ones where they'd be able to run their SoCs at higher power. They're also markets where there are advantages to going that route. It further bolsters the iPad Pro as a pro device (where it will be the ultraportable device for developers). And the Macbook and Macbook Air would benefit from the efficiency and power - especially graphics - of Apple's SoC, and the people buying them likely would be fine using Apps. From there, the Mac Mini and Apple TV I think converge (into a set top box, with it possibly even being touted as a gaming console). I think the normal Mac Mini will continue for awhile, but we'll see the Apple TV become more robust (well I could see different versions of it, with an even smaller streaming stick version, but then a more premium one - and I have a hunch it'll end up being like a shrunk down Xbox One with Kinect, where it'll have voice and gesture commands and can do motion tracking/face time/etc, but it'll be a small little box more like the Mac Mini).

I don't think the Macbook Pro does until they do a really robust performance oriented SoC. I'd guess we'll see tiers of them. And I think the Mac Pro will eventually follow suit, where basically they just stuff several of the performance oriented SoCs into it. And I think that could potentially enable them to offer server setup with like mini-racks where you just slot in a card with a SoC that has everything contained on it - really they already have the power and battery backup right there as well.