Analyst: Apple Will Switch Away From Intel in 2020

iMac

Every now and then, a rumor pops up suggesting Apple will switch away from Intel microprocessors at some point in the relatively near future, in favor of its own ARM designs. It’s a topic my colleague Grant Brunner addressed as far back as 2012, and that I wrote about in 2014 and 2017. It’s an idea being floated once again — this time by an analyst with a proven track record of being right about Apple’s long-term product plans.

According to an analyst note from Ming-Chi Kuo, we’ll see Apple deploy its own silicon in its computers in 2020, with an autonomous Apple Car system possibly hitting the road by 2023. In the past, I’ve argued that this was highly unlikely. Switching away from Intel means switching away from x86 compatibility under Windows. It potentially means taking on some extremely difficult engineering when it comes to building high-end multi-core microprocessors. Apple has a great deal of experience building small mobile devices with up to six cores, but the company has never tackled anything like the 16-32 core chips that both AMD and Intel have experience building. And while pundits are quite correct that Apple has been willing to switch CPU architectures before, the company has typically only done so when it had a significant performance advantage to gain — one large enough to wipe out most, if not all, of the performance loss associated with emulation. When Apple switched to the PowerPC architecture, the emulated programs often ran more quickly than they had on native 68k CPUs.

Some of these issues are still in play, but others have changed significantly since we started discussing this idea six years ago. Let’s step through them:

Apple’s SoC Team Has Grown Up

When I discussed this idea in 2014, Apple’s highest-end SoC was the A8 — a 20nm dual-core CPU that was already setting performance records relative to its competition, but also one that lacked the multi-core capabilities and associated complexity that other ARM customers were fielding. This was still somewhat the case in early 2017 when the then-new A10 was a quad-core chip with Apple’s first high-efficiency and high-performance cores. Today, the company has stepped up to fielding a six core design, with a high-efficiency quad-core and a pair of high-performance CPUs. It now designs its own GPU, where it previously licensed that technology from Imagination Technologies, and its added an NPU (or Neural Engine if you prefer Apple’s nomenclature). I’m not a fan of using transistor counts to compare microprocessor features, but as a measure of design complexity the metric functions well enough. Four years ago, the Apple A8 tipped the scale at 2B transistors. Today, the Apple A12 Bionic is a 6.9B transistor part.

Apple-A8-chipworks

Apple has come a long way from the A8. Image by Chipworks

No, that’s still not the same as building a 16-32 core CPU — but Apple has continued to advance its own engineering and integration efforts.

Intel Has Stumbled

One potentially significant piece of the puzzle is the difficulty Intel has had with both its 14nm and 10nm roadmaps. Apple has good reason to be wary when its manufacturing partners run into trouble: The company’s problems with Motorola and IBM were what drove it into Intel’s arms in the first place. Intel’s 14nm roadmap was delayed, but its 10nm rollout, evaluated from its original target date of 2015 to the expected “Holidays 2019” ship date will have been delayed a full four years.

I’m not going to belabor the Intel point — we’ve discussed the problem a great deal in other stories — but it’s a potential factor. Intel’s manufacturing issues over the last six years may have encouraged Apple to begin investigating building its own high-end laptop and desktop chips, even if it wasn’t otherwise planning to do so. It typically takes a design team between 3-5 years to build a brand new CPU from scratch, implying things might have swung into high gear between 2015 – 2017. (If Apple simply intends to extend its existing architecture this timeline would be different in ways I can’t really predict).

Apple has suffered through its manufacturing partners problems before. As the world’s first trillion-dollar company, it doesn’t need to suffer them any more.

Windows Compatibility Is Better…ish?

Windows 10’s support for ARM is another potential piece of this puzzle. After the debacle of Windows RT, the chances Microsoft would build an ARM version of its OS with strong emulator support seemed dim, to put it lightly. Nonetheless, Microsoft built the product. Is it good enough to replace x86 CPUs today? Flatly no. The emulation is 32-bit only, performance is highly variable, the Snapdragon 835 isn’t a desktop-class CPU in the first place, and good luck running any application requiring actual horsepower. But Microsoft could have another 18-24 months to work on its emulation software, and after seeing how well the company pulled off Xbox 360 emulation on the Xbox One, I genuinely wouldn’t want to bet against its ARM – x86 team.

None of this is to suggest that ARM performance in x86 emulation is going to match native x86. It isn’t. It can’t. The old days, when a new CPU could hit 3x the clock speed of its predecessor are long gone. But it is possible that Microsoft’s Windows on ARM emulation is expected to be robust enough to provide a reasonable compatibility layer for the Mac users still tied to the Windows ecosystem. Companies like Adobe may take six years to release native software versions, but they’ll eventually fall in line.

Are there reasons to question the narrative? Yes. Systems like the iMac Pro and Mac Pro could still be sticking points. It typically takes Intel longer to roll out high-core CPUs on a new architecture because of the increased complexity inherent to building large multi-core chips — and no, you don’t avoid all this work if you use an MCM like AMD — you just shift some of it over into the interconnect. Either the company has to ramp a very broad swath of CPUs at once, or it has to divide its own market between x86 chips and ARM CPUs. That’s a potential recipe for confusion, especially if Apple were to try and introduce one family of chips in laptops while keeping Intel in some desktops. But given the way Apple products and the larger semiconductor ecosystem have both continued to evolve, the idea of Apple introducing its own processors is more likely than ever.

Now Read: The threat of Macs switching from x86 to ARM, Why Apple won’t dump Intel x86 for its own ARM chips in MacBooks and the Mac Pro, and Apple may integrate new custom ARM cores into future Macs, but x86 swap still unlikely