I’m going to deviate from my usual topics a bit.
It’s been a rumour for years: Apple will use Intel processors in the next Macintosh. We’re all bored with it, frankly. But recently, there’s been even more noise than usual about it. CNET keeps covering the story. Now, I’m fairly sure that CNET’s about as good at predicting the future as, say, Sollog, but maybe there’s some truth it it.
The underlying assumption seems to be that Apple would switch to the IA-32/x86 architecture. That could be true, but I find it unlikely. Applications would have to be recompiled, which would be a major blow to Mac OS X. So let’s consider an alternative: Apple switches to Intel, but does not move away from the PowerPC architecture.
How is this possible? Well, Intel starts to manufacture PowerPC chips.
Yeah, you read that right. I think it makes sense.
First, from Intel’s perspective:
The PowerPC architecture is gaining serious momentum lately. It’s always been at least somewhat popular in the embedded world. The chips are simple and thus cheap while at the same time being easier to program. RISC is here to stay, and PowerPC is probably one of the chosen platforms for it for a while.
But lately, it’s been catching fire. With the Xbox 360, Microsoft has dropped Intel for a custom IBM PowerPC chip. The PS3 is also using a PowerPC chip. Nintendo, meanwhile, is already using the PowerPC. And just like that, Intel has no part of the next gaming revolution (at least on the CPU front). That’s got to smart for Intel.
Looking forward, Intel needs RISC. They had a kick with the IA-64/Itanium, and it has mostly failed in favor of AMD64, which is really a 64-bit superset of IA-32. Ultimately, nobody wanted a brand new architecture that hadn’t already been demonstrated capable. Or, maybe more accurately, nobody wanted to make the first moves.
But there’s already a RISC chipset out there gaining serious traction: The PowerPC. And what better way to get in on the market than scooping Apple from IBM? That’s a million CPUs per quarter. That’s a million more than IA-64 had on day zero. Not a bad start.
This really makes sense from Intel’s perspective.
From Apple’s perspective:
IBM has been over promising and under delivering for years. Intel, on the other hand, has been squeezing absolutely remarkable performance gains out of an aging technology for over a decade. I know that I never expected the Pentium to last this long, and I’m not alone.
Intel’s proven they can make very difficult gains. It stands to reason they can make the simple ones as well.
I’m not really expecting an announcement about a vendor change on Monday. But even if there is one, I’m expecting Apple to remain with the PowerPC architecture. And I actually hope this does happen.
To bring this back to the usual topic of my blog: We probably won’t see it from palmOne for years, but it wouldn’t surprise me at all to see Palm OS on Linux being deployed to the PowerPC for embedded systems by some licensees. (And Intel would get a shot at that, too, if they had a suitable PowerPC. It would have to be a different PowerPC than runs the Macintosh, but a bit of experience now would help.)
On a less serious note, Apple needs a new M as well. AIF (Apple, Intel, Freescale) just doesn’t have the oomph that AIM did. Can someone wake up marketing? We’ve got a crisis here.
Am I crazy? Tell me what you think.
June 4th, 2005 at 3:22 pm
Interesting idea… I know current Pentium IV and Pentium-M chips are implemented with a decoder front-end that converts x86 instructions to a simpler RISC-based form for scheduling and execution. I wonder how difficult it would be to add a PowerPC front end to that execution unit, giving a chip that can run both x86 and PPC code depending on the CPU mode.
June 4th, 2005 at 3:30 pm
It’s definitely a strange theory, but I think it’s a little less strange than Apple’s new strategy being “Mac OS X on x86 next year, and you’ll all come with us, right?” That sounds like taking the worst of Adam Osborne’s strategy (”what we’ve got now is crap, but wait to next year”) with something uniquely Apple (”we told you what we’re switching to was crap last year, but now it’s better!”).
Hmm, the idea of a chip that could run both instruction sets is really intriguing. Any idea what portion of the P4/PM is dedicated to x86 decoding?
(Btw, sorry for the article editing. I’m switching the article to MarkDown instead of HTML behind the scenes. I really hate the a href=”blahblah” link title /a syntax, and I can’t believe how smoothly it’s going.)
June 4th, 2005 at 3:39 pm
Come to think of it, this whole idea of decoding personalities has been around for a while, hasn’t it? I seem to remember reading that this would change the world in the days when the Powerbook Duo 230 in 1992 or so, and that’s only because that’s when I started really stuff by the really clever monkeys.
It could be that it makes sense to not only standardize the scheduling/execution units, but to just tack on decoding units. Want to compete with the ARM? Throw in an ARM decoder. Want to complete with a PowerPC? Throw one of them in, too. x86? Here ya go, we got one of those too. All in the same chip. I know these percentages are insane, but of the decoding unit is 60%, it might not make any sense at all. 1% and it probably would. I wonder where would it stop making sense, and if we’re above or below that right now.
There’s other implications here, too. The x86 decoding unit fails testing but the PowerPC one doesn’t? Ship it as a PowerPC-only chip.
Huh. Thanks for the thoughts, Ben.
June 5th, 2005 at 1:23 am
Well, in a sense, this is what Transmeta was doing — they had a sophisticated VLIW execution engine and code that translated x86 to it. ARM chips also use this concept; the Thumb instruction set is a front-end for their 32-bit instruction where single 16-bit-long instructions get expanded into the full size ones. I can’t say this is what Intel would be doing, but it’s certainly possible.