Apple Moving Macs to ARM? If History is Any Guide, That’s…Entirely Plausible
By Harry McCracken | Friday, May 6, 2011 at 3:30 pm
A Web site with the wonderful name SemiAccurate is reporting that it’s a “done deal” that Apple will dump Intel chips for ones based on the ARM architecture used in most smartphones and tablets, including the iPhone and iPad–and it’ll do it “as soon as possible.” I tend to be skeptical about rumors of great big news that come from not-so-well-known sites. But I’m nowhere near as skeptical as VentureBeat’s Devindra Hardawar:
Seems to me that there are several factors that make a Mac move to ARM plausible, or least very far from unthinkable…
1) We know for a fact that Apple likes to be as deeply involved in the technology inside its devices as possible–that’s why it uses custom processors based on ARM’s technology inside the iPhone and iPad. Intel’s upcoming chips based on “3D” technology sound neat, but they’ll always be Intel chips, not Apple chips.
2) By building its own custom ARM-based chips, it’s at least possible that Apple could make a great leap forward in laptop battery life compared to the rest of the industry. (The iPad already has strikingly better battery performance than other tablets based on non-custom ARM processors.)
3) OS X already runs magnificently on ARM chips–in the form of iOS, which is a variant of OS X as it runs on Macs, not an unrelated operating system. Bringing full-blooded OS X to ARM wouldn’t be an entirely new challenge.
4) Apple has already said it plans to make future Macs more like the iPad. Wouldn’t basing them on the same chips help?
5) Mightn’t the company prefer to build software for one chip platform (ARM) rather than two (ARM and Intel)?
There’s a sixth reason why ARM Macs feel like a real possibility to me–and that’s history. Especially the history of an uncannily similar rumor from a little under six years ago.
On May 23th, 2005, the Wall Street Journal reported that Apple was considering moving from IBM’s PowerPC chips to Intel ones. The experts quickly chimed in, and said that this rumor was…well, wrong. Or at least almost certainly wrong.
A couple of weeks later, both the Journal and Cnet News.com reported as fact that Apple was about to announce at its WWDC event that it was moving to Intel. This time, with two respected news sources reporting that it was definitively going to happen, the response from multiple experts was still that the idea was far-fetched.
Even Daring Fireball’s John Gruber, who has a better track record of channeling Apple’s thought process than any outsider, couldn’t figure out how to make it make sense.
And then the far-fetched, unlikely, unthinkable scenario turned out to be true:
In other words, rational analysis by knowledgable people turned out to be a lousy way of determining whether Apple was going to bet the Mac platform on a huge transition to a new architecture. Actually, rational analysis by knowledgable people almost always seems to be a lousy way of predicting what Apple is going to do. I’m keeping that in mind as I read about the ARM scuttlebutt, and think about it.
Footnote: After the Apple-Intel news became official, some analysts still thought it made no sense;:
While we can see why moving to a dual architecture approach may bring some benefits, a wholesale move away from the IBM chips would be extremely foolish. Intel is not the ‘de-facto leader in processor design’ that it was a few years ago; in the recent past, Intel has been out-innovated by both AMD (with a better approach to 64-bit computing) and IBM (with a better long-term strategy around multicore chips.
You be the judge of whether Apple knew what it was doing in 2005–and might know what it’s doing in 2011 if it transitions Macs to ARM processors.
24 Comments
Read more:
24 Comments For This Post
May 6th, 2011 at 3:55 pm
I have to disagree with you on this one. The people who doubted the Apple move to Intel in 2005 had no idea of what they were talking about. IBM had made it quite clear that it had no interest in further development of Power for anything but servers and Apple had no real choice.
Why the move to ARM makes no sense:
–If Intel can deliver on the promise of Ivy Bridge, x86 becomes far more competitive with ARM in power efficiency. Admittedly, Intel has a history of overpromising and underdelivering in this area, but the new process looks like a genuine breakthrough. If it is, it is more likely to get x86 into tablets than ARM into laptops.
–ARM might make sense in the Air. But Intel will have a ULV Sandy Bridge part soon. and it would be a disaster to use different processor types in different members of the family. And it is really hard to imagine MacBook Pros, iMacs, and Mac Pro going to ARM because they would take a vast hit in processor power.
–Apropos of that last point, it has been barely five years since Apple put developers through the wringer of having to rewrite all their code for Intel. For big complex packages like Adobe CS and Microsoft Office, not to mention more exotic media production software, this is a big, expensive deal.
There's been a lot of talk about ARM replacing x86 on servers, but nearly all of it has been just that–talk. And servers make a lot more sense because many server tasks are I/O bound and put very little stress on the CPU.
Except for the Air, the Mac market is very performance conscious (and I suspect most Air owners have another Mac around; I have both a PowerBook and an iMac.) ARM just doesn't make sense for this market.
May 6th, 2011 at 5:59 pm
Good points all. I should clarify that I don’t buy the notion of it happening “as soon as possible.” But the notion of Macs on ARM, sooner or later, in one form or another, is just as likely as…well, as Windows on ARM. And that IS happening.
–Harry
May 7th, 2011 at 9:43 am
Agreed. It won’t be the entire product line at once—these transitions are always multi-year affairs—but it will start happening in some form.
I wouldn’t at all be surprised to learn that there are a few ARM-based MacBook Air notebooks inside of Apple right now…
May 7th, 2011 at 1:41 pm
Yes as in a MacBook Air that is dual mode IOS / OS X.
May 7th, 2011 at 3:57 pm
Even with Ivy Bridge, Intel is not even close to matching ARM in low-power. And there is no reason that Intel couldn't fab Apple's ARM-based chip at 22nm tri-gate also. Thunderbolt even makes this likely, because Apple will want Thunderbolt in iOS devices and Intel wants into mobile.
ARM does not mean low-performance, just low-power. A 64-bit chip with 64 ARM cores on it is coming in a couple of years. 63 of those cores could sleep when the system is mostly idle, and all 64 could run and would easily beat an Intel chip of the same die size, which might only have 8 cores. ARM cores are almost microscopic, you can fit many more on the same size die. Die size is what costs money. So we are not talking about 1 ARM core versus 1 Intel core, we're talking about 2 chips of the same size, same price, same thermal envelope. The ARM-based one will have many more cores than the Intel-based chip. Today, Apple's 2-core A5 is smaller and faster than Intel's single-core Atom. And cheaper.
Also, if Intel is fabbing Apple's SoC's, why not have 32 ARM cores, 4 Intel cores, GPU, Thunderbolt, if that is what gives you the best power and performance and compatibility?
And part of what made the PowerPC to Intel switch hard was a) Carbon API, which has now been phased out, x64 apps are all Cocoa, b) not all developers were using Apple's tools on PowerPC, but on Intel they all are, c) there was no Mac App Store to push recompiled updates to users. A switch to ARM would be a lot easier today.
May 7th, 2011 at 4:15 pm
One prediction: ARM (and Intel too) will have 64-core chips long before Apple or anyone else has a commercial operating system that can use them efficiently in desktops, laptops, or tablets.
May 6th, 2011 at 6:53 pm
Windows on ARM is largely defensive move. Microsoft doesn't want to cede the ARM server market, whatever it may be entirely to Linux. And unless Intel delivers something really good in low-power x86, the Windows 8 tablets will have to run on ARM (the battery life of Windows 7 x86 tablets is almost as good an argument against them as the OS.)
Apple doesn't face similar imperatives driving ARM on the Mac.
Steve
May 7th, 2011 at 1:21 am
For now, I really disagree. ARM just CAN'T deliver the kind of power x86 currently delivers. It's not a matter of throwing a few extra watts at the current ARM chips to get them to deliver the pwoer x86 does: the current generation is just not aimed at delivering desktop-class performance.
Technical explanation:
To come anywhere close to x86, ARM has to move to OOP (out of order processing), which greatly increases the complexity and consumption of a chip, but it especially increases performance.
Even more than OOP, the reason x86 delivers so much more performance/clock than ARM, is because ARM is RISC(reduced instruction set computing) and Intel is CISC (complex instruction set computing). A CISC architecture offers complex instructions (With sandy bridge Intel even introduced instructions with three operands, allowing for example c=a+b to happen in one instruction instead of two) which are heavily optimized by a complex, while RISC only offers basic instructions. these are also optimized, but since doing one complex job (for example a SIMD (single instruction multiple data) instruction, which are being used more and more these days) is so heavily optimized in a CISC CPU, it'll always be faster than in a RISC CPU.
Bottom line: Intel is not stupid: the reason ARM currently has a better performance/Watt value, is because for lower performance, it's easier to keep that value high. At doing simple thing, ARM (and other simpler architectures for that matter) will always stay more efficient thanx86, but for desktop use which requires more power, ARM would be a nightmare in terms of power efficiency.
May 7th, 2011 at 10:10 am
>>ARM has to move to OOP (out of order processing), which greatly increases the complexity and consumption of a chip
A9 already does OOP in hardware, the architecture is almost superscalar like. On top of that the compilers also do instruction reordering
You are worng when you categorically say that CISC performs better than RISC, performance depends on lots of other factors like clocking, object code optimization, pipeline execution, memory latency, compiler optimization level. I felt that you are randomly throwiing terms without giving a coherent argument, it is valid to compare processor architectures with proper benchmarking data not by making proclamations like CISC is greater than RISC etc.
May 8th, 2011 at 6:21 am
And to add to Mahesh's comment that newer ARM chips can have similar performance as Intel, compilers barely ever make good use of SIMD instructions, partly because Intel's MMX & SSE units share the floating-point registers, so it is extremely rare to see significant gains (eg: 5x performance or better) by using SIMD on Intel chips. Whereas the ARM architecture is designed for simplicity, and their NEON SIMD unit is significantly better than Intel's MMX & SSE. Sure, Intel has just introduced AVX in addition to MMX & SSE, but its still not as good as ARM's SIMD unit that is several years older. And either way, SIMD is unlikely to make a significant boost unless if you use hand-optimized Assembly code, since even the most sophisticated C compilers still have too much trouble using SIMD properly.
May 7th, 2011 at 2:48 am
Step 1: Move all the apps to Mac App Store
Step 2: Make Mac App Store developers recompile for ARM
Much easier than previous transitions from 68K to PPC to x86.
Also ARM doesn't need do out of order execution because instructions can be conditional so you don't have to branch around everywhere like CISC. Or at least that's the way it was back in the early 90s when I last programmed ARM assembly. 🙂
May 7th, 2011 at 4:14 am
I agree with Harry — this is all quite possible. OSX on ARM also eliminates all possibilities of hackintoshing the OS. It also restores the mystique Apple enjoyed when they had their own G4, G5 CPUs. And it gives the company more control.
Going with Intel meant that Apple was competing head to head (hardware-wise) with Dell and HP et al, and we all know when to expect Mac upgrades — when new Intel chips come along. By contrast, nobody outside Apple knows when the A6 or A7 chips will come.
Finally, the argument that ‘the last architecture transition was not long ago enough’ falls apart when you look at what happened during the PPC to Intel switch: Apple (which at the time announced ‘we are now hardware agnostic’) at the time forced developers to use XCode and the other Apple dev tools; only the companies (like Adobe) who had stuck with the older tools were caught in difficult upgrades. Lesson learned: everybody today who uses Apple’s coding tools can be issued a simple update, then recompile to the new architecture. Apple knows how to do this better than anybody else, thanks to having done it twice before.
And the ARM Cortex a-15 ‘Eagle’ chips will be OOP IIRC.
Against all this is the wild card of what Intel with its giant fabs, billions of cash, and thousands of genius engineers, can do. Intel wants to gain market dominance in smartphones and tablets as it has in servers and PCs. They are at least one fab die-shrink ahead of the ARM licensees, and now they have this tri-gate ‘3D’ processing on top of that.
Apple will go with the architecture that gives them the most power per watt, with the added proviso that Apple will always prefer their own custom chips over any third party, generic chips that Apple competitors can buy.
May 7th, 2011 at 6:20 am
Apple makes about 5x larger profit margins http://armdevices.net/2011/05/06/apple-to-obvious… on the ARM Powered iPhone than they do on the Intel powered Macbook Air. It is the most important thing for Apple to find a next cash cow, a next product to sustain its required profit stream so that the Apple stock can stay valued so high in a market where their iPhone is being out-competed with Android in the low-cost and high-end segment.
There are a list of technical advantages of using ARM:
– With a transflective type screen like Pixel Qi, the ARM Powered Macbook could run for over 30 hours on a small battery.
– The design of an ARM Powered Macbook Air can be the lightest, thinnest yet, Apple sells that kind of design feature.
– ARM is about to be powerful enough to run a Desktop/Laptop sized basic OS, with big focus on HTML5 features and support, and the app store approach for getting apps that can take advantage of native code, everything is under control. Even the most advanced apps like Final Cut, Photoshop, those CAN be made to work through the newest types of HTML5 web apps, with full but managed native code support, with offline support, and combining those professional apps with cloud computing, you suddenly can do all the hardcore renderings on the cloud through grid computing, which basically can give every user an on-demand access to a super computer. That basically means this solution is also better for the most demanding multimedia professionals.
May 7th, 2011 at 5:01 pm
HTML5 has nothing to do with it. Cocoa is also processor-independent. Even Final Cut 1.0 cannot be made to run in HTML5 yet, and likely not for 10 years or more. If Apple puts ARM in the Mac, they will simply add an "ARM" architecture checkbox to the Mac part of Xcode and developers will recompile and push their update through App Store. Xcode already compiles iOS and its apps for both ARM and Intel, and Mac OS apps for Intel and PowerPC.
HTML5 will always be behind Cocoa, because standardization is slow. HTML5 is free with whatever device you buy, but Cocoa is what you pay for when you buy an Apple device. They will coexist going forward as they have always done. The first Web browser and server were written in Cocoa, only 2 years after Cocoa shipped.
May 7th, 2011 at 6:35 am
Sounds like a pretty good idea in theory
http://www.anon-toolz.at.tc
May 7th, 2011 at 6:37 am
What is needed is to leave X86 behind. Both Apple and Microsoft should work together to make this happen ASAP. ARM is the way to do it.
May 7th, 2011 at 8:07 am
Inte’s 3D technology is nice and a definite breakthrough, but the brealthrough is in being able to continue Moore’s Law, even if they can’t shrink the transistor’s anymore. Keyword here is “continue Moore’s law”. It doesn’t have much to do with a huge increase in performance or a huge decrease in energy consumption. This will merely allow them to keep doubling the performance of their chips every 2 years. Nothing more, nothing less.
So why would this threaten ARM’s path? It doesn’t. It just ensures Intel can continue to improve its chips. ARM doesn’t need 3D chips yet, because they aren’t yet to that point (20nm and bellow) where they need this kind of technology to continue improving their chips. And if case people haven’t paid attention, for now at least, ARM chips are doubling in performance faster than Intel can. ARM chips double in performance every 12 months, as opposed to 18 or 24 months for Intel and AMD, while maintaining the same low (<1W) energy consumption.
If Apples moves to ARM for their low-end notebooks in 2013, that could mean 2 things. First if they choose the normal mobile path for ARM chip progress, they could have at least a 2.5 Ghz quad core Cortex A15 chip by then. I think Nvidia might even have an 8 core one as Tegra 5 in 2013. Perhaps Apple could do it, too. This would mean the Apple A7 chip will be in both iPhone, iPad and say Macbook Air in 2013. But this also means they will be about as powerful, so we'll see if Apple wants to do this by keeping the same chip universally across all their devices.
Another path they could take, is what Nvidia is doing with Project Denver, which won't follow the same natural progression path for mobile ARM chips, but it will try to jump ahead 1 or 2 generations. Nvidia's Project Denver is also supposed to be out in 2013, and I think it's a custom chip which means it could even be a 3 Ghz 8 or 16 core chip, developed specifically for high-end for servers, PC's or high-end notebooks.
If Apple's design team can do something similar then they could use the quad core Cortex A15 for A7 for iPhone and iPad, and use another higher performance custom design chip, that would be similar to Nvidia's Project Denver, and would be at least 1 or 2 generations ahead of the Apple A7 chip, which could be used in the Macbook Air. Since it won't follow the natural progression of mobile ARM chips, that means it will use more power, but it will only be 2 or 3x bigger – say a 2-3W power draw. Today's Atom's use several times more than that. And even if Intel switches Atom to the 22nm Ivy Bridge in 2013( they can't do it sooner than that – only 32nm in 2012), it would still have to be the single core 1.5 Ghz Atom they are using today. Adding more cores would cancel out their efforts to get Atom down to under 1W.
So Apple would have to choose at best between an 8 or 16 core chip at 2.5 Ghz or a 1-2 core Atom chip at 1.5 Ghz in 2013. I think the choice is obvious. And don't underestimate how much Apple wants to control their own hardware. If they can make notebooks with their own ARM chips, that have "good enough" performance (doesn't have to beat Intel at all) and stellar battery life, you better believe they will do it.
May 7th, 2011 at 4:53 pm
I don't think they have to sacrifice performance. A dirty secret is your Intel chip runs slower to save power, or turns off cores to save power. So performance by itself is meaningless, it has to be performance-per-watt. If they made an ARM chip that was as big and power hungry as an i5, it would outperform the i5, whether running at peak or idling to preserve power. The trick is, the ARM cores are much, much smaller, so you would have to put more on there.
A preview is to compare A5 to Atom. A5 is smaller, cheaper, faster, and uses much less power than Atom. A5 has much more performance-per-watt.
May 7th, 2011 at 9:22 am
I think this is very plausible. Another big reason for the switch to Intel was portability with the predominant OS to remove barriers to switching. This can be accomplished on ARM with Win 8 and beyond moving to it. The reasons mentioned above are good. This is yet another reason why this is entirely plausible. People said Apple moving to Intel was impossible too. Well, here we are 5 years later and look how that worked out.
May 7th, 2011 at 11:51 am
The reason Apple moved from PowerPC to Intel was because PowerPC was lagging far behind in performance. ARM can't touch the performance of Intel chips.
HOWEVER…
Apple may be pushing portable computing into a new direction. They may have decided that computers today are powerful enough for the future of portable computing… mostly web based. They like how they have the iOS app store locked down and are trying to do the same thing on OSX. I can see their notebooks moving to a hybrid of OSX and iOS where there is a closed environment with current day processor power, but with much smaller profiles and much longer battery life. That is pretty believable to me.
They'll still need the full OSX on Intel for their MacPro hardware, though. Apple has carved out a couple of nice niches for itself… the mobile portable internet/app appliance, and the uber powerful server hardware, polycore, RAM glut, near super-computer machines for things like video editing. I can see them pushing their laptop line into the appliance category. I can't see them abandoning the MacPros. And that would leave the iMac in a weird limbo. It doesn't fit neatly into either category.
May 7th, 2011 at 4:43 pm
No, it was not performance, it was "performance-per-watt." ARM is way ahead of Intel in performance-per-watt.
You think ARM has low performance compared to Intel because right now, the only Intel chip ARM competes with is Atom. Competitors for Intel Core and Xeon are coming. But how does Apple's A5 ARM do compared to Intel Atom? The Apple chip is much faster, much smaller, and uses much less power. It's not even close. Because it is smaller, A5 is also cheaper. Theoretically, you could put 2 in a box instead of an Atom, so ARM is really trouncing Atom.
So when we see ARM-based chips that compete with Core and Xeon, they are likely to also be faster, smaller, and use much less power. You're thinking Xeon has much more performance per core, but a chip the size of a 16-core Xeon might have 512 ARM cores. And it would have better performance and lower power. Or at the same power, it might have 2048 cores.
There's no getting around the fact that Intel's chips are big and complicated. They got better performance-per-watt by becoming great at turning off more and more of the chip most of the time, and tri-gate continues that, but Apple is getting better performance-per-watt by just leaving most parts out, running a much smaller, simpler chip, and doing more in software.
Also, remember that A5 is not just a CPU, it's the entire system board on a chip. Only a minority of the space on A5 is CPU. There is GPU and other functions. An Apple chip that is the size of the system board in MacBook Air might have hundreds of cores and the MacBook Air might be faster and go 20 hours on a charge.
> iMac
With Thunderbolt, PCI is going away. There is no reason Mac Pro couldn't take on the iMac form factor. A lot of Mac Pros have been replaced with iMacs already. What the iMac was in 1998 (the people's Mac) is now MacBook Air 11-inch: a little less power than you might like, but in a form factor that makes it worth it, and a lack of legacy ports that ends up not mattering just a little while later. When MacBook Air gets Thunderbolt, that is even more obvious. So the next Mac Pro may very well be a fat iMac with Xeons and lots of RAM slots and 4 Thunderbolt ports and you hook up a Thunderbolt disk array and other accessories.
May 7th, 2011 at 4:07 pm
What did Steve Jobs say was the reason Apple switched from PowerPC to Intel? Performance-per-watt. ARM spanks Intel in performance-per-watt. Case closed. They will switch at some point. AC power is more important now than in 2005, not less important, because energy costs are way up and we use more computers than ever, and on batteries more than ever.
However, I would not be surprised if Intel starts fabbing Apple's chips first, so that they switch from 64-bit Intel to 64-bit Apple (with ARM cores) but the chips are still made by Intel. Apple and Intel have gotten closer since iPhone, not farther apart. Apple will need 250 million SoC's in 2015 just for the iPad, and only Intel can do that. Apple will need Thunderbolt on those SoC's, and only Intel can do that.
May 7th, 2011 at 5:58 pm
Summary:
1) It's complicated. There are many considerations. This subject is easily oversimplified.
2) This is one of Apple's many R&D projects, and things graduate from R&D when it's time.
August 24th, 2011 at 4:46 pm
I am curious if any one makes the connection with Apple, ARM and Web Servers.
A low electrical power server is a large prize, evidenced by Calxeda.
Apple discontinued it's xserve product, could it be replaced by an ARM version.
Apple has the source code for its own webserver, recompiling for ARM should be easy.
Could Apple be the first market?