Thursday, July 28, 2005

Apple's future with Intel  

Or, how I learned to stop worrying and love x86

It's been a while since I've posted. Work and family has kept me very busy lately. But since my last post the Mac world has received big news which would be impossible to let go without comment: Apple is switching the Macintosh computer to Intel processors. The keynote video stream where Steve made the announcement is still worth watching if you haven't seen it yet.

Enough time has passed that I'm going to fast-forward through the obvious comments:

  • Why switch? PowerPC clock speeds were stagnant, and temperatures and power consumption were staying too high. Fixing those problems was possible, but would have required a lot of expensive investment from Apple. And the problems wouldn't stay fixed: Apple would have to keep paying to push the development of the PowerPC, as they've been doing for years now. It finally reached the point where the price just wasn't worth it. Every other reason suggested is secondary.
  • Did I see it coming? Like a lot of people, I knew how much of the OS ran on x86 to begin with, and was well aware that it was an open possibility. Personally I figured it was inevitable that Mac OS X would run on x86 eventually. It's just happening a little sooner and in a more dramatic fashion than I had thought. I didn't anticipate the abandonment of PowerPC.
  • The leaks beforehand had all the hallmarks of being authentic and deliberate. As soon as I saw them coming out I felt that they were probably true. The leaks went to real news sources first, not the rumor sites. I believe the first mention came in the Wall Street Journal, and the WSJ's factual reporting is second to none. They would not have printed it if it hadn't been confirmed in some way. (Their opinion page is another matter.) And the Journal's Walt Mossberg seems to be a friend of Steve Jobs; he's been granted special early access to all sorts of Apple technologies and usually gives them glowing reviews. The WSJ is a friendly voice to Apple and they would be an excellent place to leak and start creating a buzz.

Now that that's all out of the way, let's move on. By the way, although I have friends at Apple and I'm probably covered by a ton of NDAs over the years, I want to emphasize that I'm not disclosing any confidential information here. This is all personal speculation based on publicly-available information.

Why Intel?

Why did Apple switch to Intel specifically? And what about the details? A lot of folks have been freaking out and wondering why Apple didn't say anything specifically about x86-64 or AMD.

First up, I'm pretty sure the cost savings to Apple will be more than just the CPU. The talk I've heard says that Intel offers bulk discounts to vendors who switch to Intel across the board, and Apple seems like a likely candidate for such a switch. Intel makes a a lot more than just CPUs, after all -- audio chips, SCCs, ethernet controllers, I/O processors, SATA controllers, and PCI chipsets, just to name a few. Apple already uses a lot of non-CPU Intel chips in Macs, Airport base stations, and iPods. It's certainly feasible that Apple might be going all-Intel-all-the-time.

Nor is it just chips, either. Remember, Intel designs motherboards too. Right now Apple does its board design in-house, which creates a significant lead time for product development as new boards are tested and reworked. Discounted Intel boards -- perhaps fully tested and qualified before they even reach Infinite Loop -- could be yet another potential cost savings. Apple has a penchant for pushing design boundaries with iMacs and laptops, true, but the company still does a brisk business selling desktop machines where space is not at a premium. Besides, Intel has been stuffing Pentiums into tiny spaces lately -- witness their recent clone of the Mac mini.

Could AMD compete with all that Intel has to offer? Frankly, it seems unlikely.

The door will still be open for AMD and other x86 chip vendors. I don't think Apple is going to start using Intel-specific features that AMD can't compete with -- that would be a foolish return to the single-chip-source problems that plagued Apple with the PowerPC. It's likely that the companies have made a deal where Apple commits to Intel for some number of years on negotiated terms, and thereafter the door is open to renegotiate or seek a better deal. So AMD might still get their foot in the door eventually. If Intel starts causing problems you bet Apple will switch to AMD or someone else; but as long as they offer a good deal they will be difficult to beat.


What about 64-bit support? Some folks were upset because Apple had done all this work toward 64-bit PowerPC and seems to now be ditching it. That work absolutely is not wasted. Anyone who thinks Apple is suddenly ditching 64-bit computing should put down the crack pipe -- it's not gonna happen. Sure, after Intel came up with the all-but-failed and incompatible IA-64/Itanium architecture, AMD turned around and created a much more popular backward-compatible architecture called x86-64. Point to AMD. But Intel at least recognized its mistake and cloned x86-64 for use in recent chips.

Apple may not make the jump to 64-bit Intel chips in its very first release, though I think it's at least possible that they might. But if they start with 32-bit chips, I will guarantee that they'll be on 64-bit chips within the year. And further, that they'll be going with the x86-64 architecture rather than the Itanium. It's the only decision that makes sense.

Should you care?

My personal opinion is that as a consumer you probably won't need to care about the transition.

Think about it: why do you buy a Macintosh? Do you really care about the chip that's inside it? No. What you want from a Mac is that you want the nice user experience, you want your apps to work, and you don't want the machine to stink up your desktop and crash and be virus-ridden like Windows.

Apple's switch to x86 will not change anything about the operating system or the applications: it will look and work exactly the same as your current Mac. It will still be just as stable and easy-to-use, all of your software will still run, all of your data will copy over and work just fine, and no viruses are going to magically jump over from Windows onto your Mac just because the hardware is the same.

Should you delay hardware purchases while you wait for the new machines? Well, you could. Apple seems to be expecting a certain number of consumers to make that choice -- they have more cash on hand than ever before ($7.5 billion USD), which it's pretty clear is a buffer against an anticipated temporary decline in Mac sales.

But the new machines won't be out for a couple of years. The first new ones might not be all that great, either; you may want to wait a few months for the second generation. Personally I'm planning on following my normal upgrade schedule: I'll keep my dual 2GHz G5 til it's on its last legs or until I get new free hardware, whichever comes first. The typical Mac lifetime is around four years (as opposed to two for PCs), and that's lengthened a bit by the way PowerPC clock speeds have been lagging, so I probably won't need to upgrade until the new machines are out anyway. As for my laptop, a late-model TiBook, it's getting pretty elderly. I wouldn't mind getting a new PowerPC laptop right now if I could afford it -- but so far furniture and family stuff has delayed me and I don't really need a new one badly enough yet.

So for now I think it's not a big deal: buy a new Mac if you want one and don't worry about it. As we get closer to the release of the first Intel Macs, though, the dropoff in Mac sales will be steeper. People will naturally wait for the new machines.

It'll be interesting to see what Jobs does with the release of the new machines: Will the release date be known or widely anticipated, for example Macworld SF 2007? Or will they get sprung on everyone unexpectedly? Several years of watching how Steve Jobs does things leads me to suspect the latter -- they will probably release the machines at least four months earlier than anyone expects. But it's an open question.

Compatibility and speed

Apple is including a way to run PowerPC apps which it is calling Rosetta. As reported by C|Net and Wired, this is Transitive's multipurpose DR emulator.

In an ironic twist, I've heard that Microsoft is also using Transitive's emulator ... but to translate x86 to PowerPC so that old games can run on the next-gen Xbox. Man. What's next -- cats and dogs living together?

I've tried out the emulator on a development system. Rosetta delivers great performance for simple apps. For a geek like me it's really cool: PowerPC apps just work transparently, and they run fast too!

However, anything that is heavily Altivec-enhanced will probably take a big hit when running under the emulator compared to running on a G5. The emulator does not emulate Altivec, so the app will first fall back to unvectorized floating-point, which could drop its speed to 25% of the vectorized floating-point. Then it would incur the cost of emulation on top of that. So some specialized operations running through the emulator might in theory be as much as 5x slower than on a comparable G5 system. As soon as the app goes native, however, it will gain all of its performance back and more.

But that only affects heavily vectorized applications. Your word processor will continue to work fine under emulation and will in fact probably be faster than before: the extra CPU speed will give it a boost. Games are mostly pumping video textures out using OpenGL, so they will probably not be affected.

Probably the most visible area where you'll see applications slow down is anything that uses QuickTime to compress or decompress video. That includes video playback; frame rates may drop significantly under some codecs. Why? Remember, Rosetta is not a mixed-mode architecture; it runs an entire app from top to bottom. And QuickTime loads and runs inside your application. So even though a native version of QuickTime will be available to native apps, an emulated app will run emulated QuickTime which will be much slower. If I've got all that right, then you can bet Apple is making an extra push to evangelize QuickTime developers to port as quickly as possible.

G4 emulation: Someone asked me the other day whether Apple might upgrade the emulator, which currently emulates a PowerPC G3, to emulate a G4 with Altivec. Frankly I doubt that will happen. It would take a big investment of time and money. And the benefit Apple would get from doing that is only in the short term, during the transition period where you want all the extra speed possible out of the emulator. After a while CPU speeds will increase so much that it just won't matter. The ROI (return on investment) just isn't there; too much work and too little benefit.


Apple has not announced plans for making the Classic environment run on the Intel machines. As a practical matter the current release of x86 Tiger does not support Classic. But have you noticed that they have not vocally announced that Classic is dead, either?

My guess is that Apple may be working on porting Classic, unannounced. There's a not-insignificant minority of Mac software that still needs it. And Apple has a long history of unbroken compatibility which it would be a shame to end.

In a lot of ways Classic is just a normal application: a lot of hardware-specific details were abstracted out of OS9 during its final years. The emulator doesn't need to be ported, after all: there's no reason why you couldn't run Classic's 68K-to-PPC emulator inside the new PPC-to-x86 emulator.

Of course, in other ways Classic is still very 'special'; there are many hacks all over xnu for it that give it special access to supervisor-level PowerPC stuff which Rosetta isn't going to emulate. Still, it seems like it'd be well worth Apple's time to sic a couple of guys on it for a year or two -- the potential gain is huge.

I suspect there are probably some doubts about whether it's technically possible. That's why there hasn't been a big announcement one way or another. But I'll go out on a limb and say there's a good chance someone will figure out a way to make it work.

Windows emulation

Here's an interesting one. Running x86 will make it a lot easier to run Windows applications on your Mac. Things like Virtual PC will run just about at full speed. Of course, Microsoft might not be very interested in porting Virtual PC to the new architecture, since Mac OS X just got a lot closer to becoming a competitor to Windows. if past history is any indication they'll probably do it eventually, but I bet they will drag their heels.

Other options exist, of course, such as VMware and WINE. It seems very likely that VMware is working on porting their software to x86 Tiger right now, and DarWINE is already underway.

Personally I think this would be an interesting thing for Apple to investigate. They might choose to be hands-off and just bundle VMware, Virtual PC, or DarWINE, but all of these are a little clunkier than what Apple prefers. It's possible that Apple might choose to enter the market themselves and deliver integrated Win32 emulation in some way.

Dual-booting into Windows: This sort of falls under this category. As others have noted, in addition to virtual PCs, it will be possible to dual-boot or triple-boot your Mac. Darwin can easily support the partition map styles used on Windows, so there's no reason you couldn't have a Linux partition, a Windows partition, and a Mac OS X partition. Mac OS X's built-in BSD and X11 are so good, however, that there's probably not a lot of reason for anyone who's not a Linux developer to dual-boot into Linux.

HP, Dell, and more

What about all the talk about whether Apple could or should release Mac OS X in a general release to other PC makers, like HP and Dell? Some people think that would've been a better idea. This attitude is exemplified by the somewhat goofy article Apple's Colossal Disappointment which was posted to Slashdot recently.

I call the article goofy because the author pays no attention whatsoever to business realities, nor does he seem to grasp the concept that policies may change over time. His basic complaint is that Apple is limiting its OS to run on its own hardware for the time being, and he thinks that's a mistake. (Or a "colossal disappointment", to use his exaggerated phrase.)

Here's my take. There were three possible ways Apple could have used its x86 code.

  1. Switch the Mac hardware to x86

  2. Allow a select few PC vendors to ship hardware that runs Mac OS X.

  3. Release Mac OS X widely for anyone and everyone with a PC, the way Microsoft releases Windows.

Apple started with item one for the reasons mentioned above -- the PowerPC has been weighed in the balance and found wanting. But the second two courses of action are not immediately feasible, nor are they wise business decisions.

Sure, they are potentially desirable goals that are at least on Apple's radar. I guarantee you Steve Jobs is thinking about them; when interviewed by Fortune magazine he mentioned that three of the biggest PC makers have repeatedly asked him to license Mac OS X for them to bundle. (My guess: the big three are HP/Compaq, Dell, and Sony.)

But it makes absolutely no sense to do all of those things at once.

What do you think would happen if Apple switched to x86 and immediately let others release x86 machines with Mac OS X too? It would undercut Mac hardware sales in a big way. Apple has been burned by cloning before in the 1990's. The circumstances are different enough now that they may try it again, but they will definitely start out slowly and carefully. It will be several years before you see anyone but Apple shipping Mac OS X.

How about making a widespread general release? First of all, you have the same problem as above where Mac hardware sales are undercut. But okay, perhaps you might make that up with increased software sales. No big whoop. But the real cost comes in the support infrastructure. Mac hardware is fairly dependable and predictable, and Apple does a good job with the small amount of tech support that is needed. But making a widespread release would suddenly skyrocket the number of support calls being made -- people would buy it for their JalopyTech 3000 computer and something would break. Apple gets a call. Frankly, Apple doesn't have the support infrastructure to handle a twenty-fold increase in phone calls. It might be possible to get there in the long run, but it won't happen overnight.

The correct course of action from a business perspective is to do exactly what Apple is doing -- switch their own machines only at first. The other things will still remain possible future directions, and I think that both are likely in the long run. But Apple is an "old" company by tech industry standards, and doesn't plan on biting off more than it can chew at any one time.

The Lockdown

Will Apple use LaGrande as part of a scheme to make sure Mac OS X is running on Mac hardware? Maybe. But it seems very likely to me that Apple won't try TOO hard. They want to discourage the casual user from running an unsupported configuration, and they will. But it's not worth the effort (and would frankly be counterproductive) to lock out the hardcore geeks.

A small amount of 'geek piracy' will cost Apple practically nothing in hardware sales. Contrary to their huge presence on the internet, the actual real-world market share of geeks is fairly small. The vast majority of Apple's and Microsoft's markets are people who wouldn't know how to perform that sort of hack, and wouldn't want to, because they know they wouldn't get support for it. Overall it's far simpler to just buy a Mac in the first place.

Honestly, it's like free advertising... let the geeks hack OSX to run on their machines. Once most geeks start using OSX their prejudices evaporate and they quickly get hooked on it, talk to their friends about it, recommend it, and so on. And that will ultimately grow the Mac's marketshare.

Further Reading

A few interesting articles in case you missed them: