- The Tech Night Owl — Cutting-Edge Tech Commentary - https://www.technightowl.com -

More Mac on ARM Speculation

There’s no dispute that the lack of serious Mac refreshes this year can be blamed on Intel. The new Broadwell chip family, using 14 nanometer dies, is late to the party. The chips that Apple requires for Macs aren’t available yet, and that may push those refreshes off to 2015. So to make do, Apple is using slightly faster chips from the previous generation, Haswell, to convey the impression of a speed bump. The combination of slightly lower prices and slightly better performance makes the recent MacBook Air and MacBook Pro with Retina display upgrades pretty good values overall.

Clearly customers approve, since Mac sales growth continues to outpace the industry.

Now the last time a chip manufacturer didn’t deliver the goods to Apple, a wholesale processor switch followed. That happened in 2005, when the PowerPC G5 stagnated, with no note-book version in sight. Windows PCs with Intel Inside were speeding past Macs, so Apple went with Intel.

That transition was quite smooth, since Apple had been developing an Intel version of OS X for several years in case a move became necessary. So new Macs sold in far greater quantities, and it didn’t hurt that you could run Windows software at good speed via virtualization or at full speed via Boot Camp.

Since then, Apple has established their own processor design facility, and the 64-bit A7, based on the ARM architecture, has been a revelation. After misleading claims from some tech pundits that there was no need for 64-bit on a mobile platform yet, the performance of the iPhone 5s, iPad Air and iPad mini with Retina display proved otherwise.

But that’s a low-power chip designed to perform well in an environment where resource use has to be carefully managed. What if the A7 and its successors were scaled up to operate in a note-book or desktop personal computer? That the A7 was classified as a desktop class processor may be the hint that has helped fuel the Mac on ARM theory.

It goes like this: In the tradition of Steve Jobs, Apple wants to control everything, and building custom chips is only one part of the picture. Clearly that strategy has worked. So here we have Intel not being able to deliver the parts on time. What’s more, an ARM processor costs far less to build, in part because of design efficiencies and not being weighed down by the legacy x86 architecture that stagnates development of Intel’s chips, in part to stay compatible with older Windows apps and operating systems.

If an ARM processor can come close to the performance of an Intel chip — and that’s a huge if — Apple would still benefit from the lower design and fabrication costs. The prices of new Macs could be reduced by hundreds of dollars, particularly at the high end where some of those Intel i7 chips cost more than $500, each. Apple also has past experience moving to new processor platforms, so isn’t this a win for all concerned?

The theory sounds promising, but there’s a severe limitation that’s rarely mentioned. There are tens of thousands of existing Intel-based apps that would have to be translated or emulated in a new chip architecture. When Apple went to PowerPC, they had an emulator that was mostly compatible with existing apps, although performance suffered at first. A key part of the Intel migration was Rosetta, which allowed you to use PowerPC apps on an Intel-based Mac; it was discontinued with the release of OS X Lion in 2012. Apple clearly believed six years was long enough for you to upgrade or replace your legacy PowerPC apps, although that’s still not entirely true.

So if Apple goes to ARM, they would have to provide simple conversion tools or the ability to create “fat binary” apps that would work on Intel and ARM. There would also have to be some sort of virtualization capability so your Mac would still look and work the same as now, regardless of whether the app was built for Intel or ARM.

How would Apple accomplish that virtualization feat? Well, consider the new Metal architecture of iOS 8, which allows the CPU and GPU to “work together to achieve optimal performance.” Right now, Metal is designed to make games run much faster. But what if Metal could be tailored to provide Intel virtualization on an ARM-based Mac? Would that afford a level of performance that would be close to or identical to a native Intel chip?

Now I do not presume to know enough about the technology to claim that a Metal virtualization alternative is workable and would deliver competitive performance and superior compatibility. But it’s also clear that Apple doesn’t show its full deck of cards, and using Metal for gaming may only be the beginning.

Certainly harnessing the power of graphics chips to enhance computing is nothing new. Apple’s OpenCL technology is really exploited on the new Mac Pro, which contains workstation-class AMD FirePro graphics, to perform advanced 3D rendering and complex scientific calculations.

While I don’t know the future possibilities of moving Macs to ARM, I have little doubt that such Macs are currently being tested in the development labs. But it’s not just about control of the platform; it’s still Intel’s game to lose. If the Broadwell and future chip designs continue to run late, you can bet Apple won’t stand still.

Sounds promising, but an ARM switch wouldn’t be a cakewalk for Apple. Performance and virtualization will remain the key obstacles, and, despite some speculation on my part, I wouldn’t presume to guess how both issues will be resolved. It is also possible the stumbles in Intel’s current processor roadmap will soon be resolved, thus reducing the possibility of a move to ARM.