Charlie Demerjian of SemiAccurate posted a fascinating article with rumours that Apple may be switching from Intel’s x86 processors to their own ARM models across their laptop range as soon as 2013.

Many commenters have been highly skeptical of this, verging on the angry denial.

Suicidal

After all, surely, that would be suicidal for Apple; even today’s best ARM chips are many times slower than Intel’s Sandy Bridge in terms of raw performance and it’s incredibly hard to shift all today’s software over to a new architecture, especially if the new platform is too slow to emulate the old one. Fancy running Photoshop CS5 in an emulator? No? Me neither.

Add in the fact that there are thousands of peripherals like printers and scanners which require drivers that are not available for Mac OS on ARM; should users just dump these?

So, would it be suicidal? Well, maybe and maybe not. In response to this, I posted the following answer on Quora and thought it would be interesting to post here with some additional detail.

In short, I believe not only that it’s plausible but that it’s actually hard to find solid arguments as to why it wouldn’t happen.

Could it happen?

A change of this magnitude could really only be completed several years out from now

Today, x86 still holds a large performance lead over ARM and while Mac OS has a large base of x86-only software and slow-moving partners in terms of taking advantage of the latest opportunities made available to developers, but it seems a possible long-term strategy for Apple to follow, unifying their platform.

Naturally, with ARM having already demonstrated such products (albeit, with no compelling software to run on them) the first product would seem likely to be a thin & light MacBook Air-style device often used as a secondary computer.

But why??

Well, one needs to look at some trends and indicators to spot why this is so likely:

The trends and indicators

1. Apple have not been afraid to jump architecture before. They have the know-how to do this.
2. Mac OS X is effectively already running on ARM in the form of iOS – it’s just a difference of form factor and performance at the moment, as well as compatibility with existing software.
3. Raw CPU performance is not as critical to overall performance as it once was, with a wide variety of specialised accelerators being part of a system to do the common jobs and bigger gains coming from technologies like SSDs.
4. iPads are proving capable of doing a large swathe of computing tasks for many people, causing a drop in netbook sales.
5. Apple see iOS as more important than Mac OS (for example, iOS has an SVP; Mac OS only a VP).
6. The cost savings of moving to an in-house CPU, chipset and GPU are huge (and they’ve already done that for the iOS devices). Do not underestimate how large these are!
7. Apple bought two CPU design-related companies and now manufacture tens of millions of processors every year.
8. There have been multiple adverts for CPU microarchitects recently posted under Apple’s “Mac Hardware” jobs section, not just the iPhone or iPod divisions.
9. Much of the momentum in the industry is with ARM at present, in terms of Operating Systems, critical software (like browsers and plugins) and hardware vendors. Most of the innovation in new software is on the web or mobile, not desktop operating systems. The Mac App Store on x86 has had far less success and holds far fewer titles than iOS on ARM.
10. OS X is becoming more iOS-like from Lion onwards.
11. Connections are becoming more wireless; drivers for peripherals are becoming less important. The future is in higher-level wireless protocols like AirPlay and AirPrint which are not specific to one device.
12. Apple aim to be vertically-integrated, able to control their own destiny and able to keep hold of their advantages. Buying in the main components of their computers from Intel, nVidia and AMD are contrary to this goal and lead to fiascos like the nVidia “bumpgate” recall on MacBook Pros.
13. XCode already compiles iOS apps to both x86 (for the simulator) and ARM (for the device) so it should be relatively simple to make a well-designed app run on both architectures. OS X universal binaries could support these in one package to ease distribution, just like they did before.

But ARM is slower, right?

It’s very hard for many people to see ARM chips outside of the context of its history in mobile but, ultimately, ARM is just an instruction set. While there are reasons why one ISA may be faster, smaller, easier to code for or more power-efficient than others at a certain power level, the main differences come from the constraints imposed on the chip designers by their target markets.

Only building a few thousand chips a year for devices with excellent cooling and vast power supplies, just to achieve maximum performance? Well, you can build a massive beast of a chip like Intel’s Itanium or IBM’s Power7. But you won’t be putting it in a mobile phone. Likewise, a CPU designed for a handheld gaming device won’t be ideal as a weather-predicting supercomputer.

Yes, it’s certainly undeniable that, today, ARM-based products are well-behind Intel and AMD’s finest on absolute performance but will that always be the case?

ARM chips have mainly been slower than x86 chips because they’ve been designed to be. They’ve been aimed at tiny, mobile devices and have had power and size constraints from that world… but there’s no reason they have to be. ARM are rumoured to have targeted their CPU cores as being no larger than 2.5mm² and using less than 0.25W.

What point was there for ARM to try to build a chip for the high-performance desktop market when all the software only ran on x86? Who would have bought it? Naturally, they focused on their main market: ultra-mobile and embedded devices.

Yet, Apple’s A5 chip is already bigger (in terms of silicon area) than competing x86 products from Intel and AMD – ARM’s ecosystem is starting to break out of this box with a vengeance so preconceptions need to be re-thought.

When it last sold on the desktop (in Acorn RiscPCs) ARM chips were often just as fast as Intel ones (if not faster in cases) but on the wrong side of the river in developer land, where no software was growing. As Steve Ballmer famously said, “Developers, developers, developers”

Really, ARM ended up in mobile as much out of luck and necessity as from a deliberate strategy initially (and, somewhat coincidentally, with the assistance of Apple for its Newton project).

But, now, it’s x86 which finds itself in the position where software looks to be stagnating and most of the growth is over on the other side of the river in ARM land (or on the web, where the client CPU architecture is largely irrelevant, excepting for plug-ins).

Expect things to change.

My best guess at how it will play out is something like this:

1. Apple announce that Mac OS X Lion will support ARM processors and start seeding the market with developer tools to recompile apps for this platform.
2. 2012/3 MacBook Air launches with a 28nm quad-core A6 processor based on Cortex-A9 or A15  that outperforms the existing Core 2 Duo in today’s MacBook Air. It runs native, ARM-compiled versions of iLife and iWork. Technology journalists criticise it for having too little CPU performance (just like they did to the previous MacBook Airs) and having a limited software library. Millions are sold anyway.
3. 2013/14 Apple hits the ground running with their own 64-bit ARM-compatible CPU designed from scratch for much higher performance, comparable to a mid-range x86 mobile chip and nVidia’s Project Denver, then pushes it out across most of their laptop range.
4. 2014/15, next generation architecture comes out, maybe even with multi-processor support, delivering good enough performance for the rest of their product range, enabling the transition to complete.

That’s just one possible scenario, of course: usual disclaimers apply, and there are any number of ways Apple could do this. The transtion from PowerPC to x86 was, after all, lightning fast.

But, by doing this, not only do they gain more control, gain a unified platform for software development, and gain the ability to produce thinner machines with better battery life, but they save literally hundreds of dollars on every Mac they ship.

Looking at today’s MacBook Air, up to $300 of the Bill of Materials cost may be assigned to the CPU, chipset, graphics and associated components according to list prices. iSuppli estimate that the Apple A5 chip costs $14 so, even if it were to more than double for a faster, bigger design, you would be talking of a saving of, conservatively, $200-250.

In its first quarter (Q4, 2010), Apple were estimated to have shipped more than 1 million MacBook Airs. Moving to ARM could have earned them a cool $250 million more profit in that quarter alone. That’s a billion a year just on the Air! In Q1, 2011, they shipped nearly 3 million laptops.

Should you trust a rumour from SemiAccurate?

Rumours are just that: rumours. Some do come true, some may been supposed to come true but later the story changed and some were always plain wrong.

Charlie has a pretty good hit rate with his: at The Inquirer, he dug down into problems with nVidia’s laptop GPUs to find the evidence that, despite nVidia’s denial, new lead-free solder bumps were too rigid and were failing because of the high number of repeated thermal cycles in a mobile device.

He drew attention to the damage to the relationship between Apple and nVidia caused by this and how Apple’s response was to replace them across their range. Well, sure enough, that appears to be happening: the only Apple products using nVidia products are now the Mac Mini, plastic MacBook and MacBook Air, all of which currently use a Core 2 Duo CPU that is is firmly on its way out in the next refresh of these lines.

However, he also predicted (as did others) that iPad 2 would have a Mini DisplayPort, which proved incorrect, even though it seems likely in the long run to feature on future iOS devices in the form of ThunderBolt.

Apple’s secrecy wall is tough to crack, for very good reasons, so the sources may or may not be accurate but, even if just a piece of sensible hypothesis, this is right up there.

Regardless of whether it happens by 2013 or not, this is clearly a massive opportunity for Apple and is fully in line with their overall strategy. Anybody want to bet that it will never happen?