My prediction in 2016, was that Apple would switch to ARM based laptops in 2019, so it looks like I am off by just 1 year.
A quick summary of my argument back then was that ARM was getting fast ENOUGH, but are considerably cheaper. Standardizing on one CPU architecture for all their products makes sense. In addition it allows them to differentiate themselves from competitors.
The only choice I am still uncertain about is the Mac Pro lineup. Since performance will matter a lot more in this segment it is harder to see a transition away from intel. A possible solution would be for apple to go massively parallel and simply use a lot of ARM CPU cores. Sure single thread performance wont be able to compete with intel but they might not need to.
A lot of the stuff you need a Mac Pro for like compiling large programs fast, video processing etc is likely very paralelizable.
> Sure single thread performance wont be able to compete with intel but they might not need to.
In regards to audio applications like Logic and Ableton, single thread performance is probably the biggest factor. Individual plugins and effects are all single threaded within a DAW.
That would definitely be an interesting and very surprising development. Like I would really like to hear Apple's rational for developing an x86 CPU.
x86 is a far more complex CPU architecture which they have zero experience in making. ARM in contrast they have experience with all the way back to Newton, and it has a nice licensing preposition.
I can't see them doing something different than ARM and x86. That would be an enormous undertaking to come up with an entirely new CPU architecture.
x86 really has nothing going for it architecture wise. It just has a market because Intel makes fast chips that run existing software. ARM is a better architecture, and Apple already has the toolchain to make software run on that architecture.
If you asked two equally competent chip makers to make a fast chip, and one was given x86 architecture and the other ARM, I am certain the ones who got ARM would win. x86 is just winning today because intel is filthy rich and can pour insane amount of money on x86 to make up for all this short comings.
> x86 really has nothing going for it architecture wise. It just has a market because Intel makes fast chips that run existing software.
People buy computers to run software. If they can't run their existing software on it, the computer has no value. Just like with the original PPC-to-Intel transition, where Apple provided a PPC emulator on Intel, if they go with ARM, Apple will have to provide Intel-to-ARM translation software in macOS for several years.
But now, a significant number people are using Macs as hosts for x86 Linux and Windows virtual machines. This changes the calculus a bit. It's one thing to emulate an ISA to run software in a user context; quite another to perform cross-ISA virtualization. QEMU can do the latter, but at a significant performance penalty.
It should also be noted that Apple is sitting on _much_ more cash than Intel. At the end of last year, Intel had $14B in cash on hand, while Apple had $285B.
"People buy computers to run software. If they can't run their existing software on it"
Obviously, but my point was that this mattered more in the past. Apple has so many tools and arrangements in place today to make a transition less painful. They got fat binaries and bundles. They got bitcode. They have really strong control over the app delivery process by having it all go through their app store. Almost everybody use xCode which they also control.
I can't remember ever running very much on PPC emulation on last transition. Usually I used fat binaries. Apple has been through so many of these transitions and gotten better at it every time. This will likely be quit smooth.
"But now, a significant number people are using Macs as hosts for x86 Linux and Windows virtual machines."
Really? I've been a Mac user since OSX came out and have very rarely encountered people who do that. I think this user group is too small for Apple to care. Anyway Linux should work fine, and I believe Windows also runs on ARM although not all software so I think this is not as big of an issue as you might think.
> I've been a Mac user since OSX came out and have very rarely encountered people who do that.
Are you a software developer developing for Linux targets? That describes nearly _all_ back-end development work done in the last 10+ years in SF and Silicon Valley.
> Linux should work fine, and I believe Windows also runs on ARM although not all software
Have you actually tried running ARM virtual machines on ARM hosts for these two environments? This can hardly be described as a mature art. If even remotely usable software exists today for this purpose, I'd love to see it.
Even if it did, the fact that developers usually target x86 builds (for deployment in the server clusters) makes this less useful than one might otherwise think.
If they are not sticking with x86, it basically has to be ARM. Producing a new CPU architecture is very costly endeavor and other CPUs exist. However because they already have experience making chips for iPhone and friends and ARM is at least "good enough", it makes sense for them to go all in with ARM.
Even at $15-20B all cash, Apple would be overpaying, and then having to deal with all of the continuing business of supporting AMD, including the ATi problem - there's just no way to acquire AMD without destroying the graphics/gpu-compute market (you know, any more than AMD has already done themselves). There's probably some other explosive x86 licensing terms that would have to be hammered out under the transfer as well (iirc, in the past when AMD buyouts have been discussed, to deal with the architecture licensing stickiness they all had to be "reverse mergers" with AMD, and that killed every prospect of that happening...)
Just not worth the hassle compared to the price of an ARM license or sticking with Intel (or even just investing a huge chunk of cash in AMD to build super power efficient laptop chips, if you had any confidence that they could pull it off).
The AMD-VIA-Intel patent cross licensing triumvirate has buyout, sellout, and a bunch of other miscellaneous prevention/mitigation clauses.
Essentially you cant just buy Intel, AMD, or VIA and acquire the rights to make x86/x86_64 processors. You could acquire a controlling or other majority ownership stake in one of the companies and then continue to run them as a wholly owned subsidiary.
In addition, Intel, AMD and VIA cant just license you the rights either. However you could enter into a Joint Venture with Intel, AMD, or VIA as a ‘passive partner’ whereby the subsidiary has inherited access to the technology. This route has been undertaken without legal repercussions so far to VIA. Zhaoxin is a joint venture between VIA Technologies and the Shanghai Municipal Government, and has so far produced a line of x86 compatible processors on a 28nm process for the Chinese domestic market that are apparently performance competitive.
Apple has more than enough cash to fund a joint venture with AMD or VIA that could enable them to bypass any possible complaints from Intel about emulation violating their intellectual property. Apple could take this route to build an ARM based processor that had sufficient x86_64 helpers to speed things up enough that the translation/emulation performance was negligible. Heck they could use this to design their own entire x86_64 CPUs and if it was with AMD they could borrow a GPU core they are familiar with too.
At the end of the day, Apple has enough money they can basically do anything they want with regards to technology.
The challenge will be successful execution in the face of a market that has become increasingly immune to the reality distortion field Apple projects, the demise of Steve Jobs significantly decreased the power of Apple to distort reality in a way that is consistently in its favour. Not so much of a ‘If Steve was still at Apple’ as simply ‘Apple isn’t the same now, and this changes the mathematics of the situation’.
Thanks for the insights on the AMD/VIA/Intel triumvirate. It’s such an odd model that x86 has landed on.
> Apple could take this route to build an ARM based processor that had sufficient x86_64 helpers to speed things up enough that the translation/emulation performance was negligible
Totally agree, this is exactly what I thought when I read the story. There’s a lot of transistor budget these days for special hardware, especially in a larger / more power-hungry part that is substituting for something quite expensive (Intel chips).
Mac’s role as the heavy-lifter in Apple’s ecosystem means it has different trade-offs to consider, and my guess is that definitely points to a JIT or compatibility layer for legacy x86 apps. If that’s the case, it wouldn’t be worth the effort unless it could be reasonably performant, which means silicon. The optics would be really bad if they launch a platform transition and it was less than ~75% of last year’s computer.
With that said, they’ll be pushing hard for app developers to transition over, and might throw away that compatibility silicon after a few years. It’s a really cool thing to be your own chipmaker and have the levers that Apple does in this situation to add compatibility silicon and not just accept another architecture’s limitations.
x86_64, but of their own manufacture. A fair number of people use MBPs to run Docker containers and other Linux/Windows VMs; to ask them to sacrifice this functionality would be to risk a fairly significant user population.
If other companies could legally figure out how to license and build x86 chips, nVidia would already be there stamping them out. It's a legal minefield that's simply not worth exploring, even for Apple.
...to compete with Intel? Why would Apple want to and nVidia not want to? Apple sells computers, nVidia sells chips.
Apple loves vertical integration and all, but selling chips is the reason nVidia exists. Apple could put out a clothing line, switch to building luxury watches or cars, or commit entirely into doing "fashionable audio" (Beats, Apple Music) to compete with Bose and still exist past computers - it's a lifestyle/fashion company masquerading as a tech company. Apple making chips is means to an end - to fill a hole in the market that Qualcomm and Samsung weren't able to. If it were anything more than that, Apple would be taking customer orders for A11s.
Meanwhile, nVidia's Project Denver and the whole Tegra stuff only came about after their plans to attack Intel head-on failed miserably. They hired up a bunch of Transmeta employees in the hope they could implement a binary translating CPU that would "cheat" them into being able to implement an x86 processor without the licensing stupidity. Intel's lawyers proved them wrong.
My prediction in 2016, was that Apple would switch to ARM based laptops in 2019, so it looks like I am off by just 1 year.
A quick summary of my argument back then was that ARM was getting fast ENOUGH, but are considerably cheaper. Standardizing on one CPU architecture for all their products makes sense. In addition it allows them to differentiate themselves from competitors.
The only choice I am still uncertain about is the Mac Pro lineup. Since performance will matter a lot more in this segment it is harder to see a transition away from intel. A possible solution would be for apple to go massively parallel and simply use a lot of ARM CPU cores. Sure single thread performance wont be able to compete with intel but they might not need to.
A lot of the stuff you need a Mac Pro for like compiling large programs fast, video processing etc is likely very paralelizable.