|Thursday, June 9, 2005
15:16 - It's still impossible, but we're doing it anyway
John Gruber has posted his lengthy and detailed post-mortem of the Intel announcement, replete with discussions of why everybody (including himself) seems to have been so spectacularly wrong about it, except for pundits who have been predicting an Intel switch for so long and so dogmatically that they're only correct this time by accident.
Rosetta — the technology that allows existing PowerPC software to “just work” on Intel-based Macs — is the missing link that makes this transition possible. “Emulator” is perhaps not quite an apt description; Apple seems to prefer the term “translator”. The specific description I’ve heard is that it is “dynamic binary software translation”. I’m curious to know more about how it works, but the only important questions are whether — as it was described in a slide during the keynote — it’s “Fast (enough)”, and how many important apps run under it. We should find out soon enough, when benchmarks start leaking from seeded developers. (Their NDA forbids publishing benchmarks based on the developer transition kit hardware, but come on, you know they’re going to leak anonymously.)
Well, that didn't take long.
The Rosetta numbers aren't very stellar. Well, in some areas they are—in some places even beating the native G5 platform—but there's clearly a lot of room for improvement. The biggest one, naturally, is Altivec/SSE# stuff, which Rosetta ostensibly doesn't support, although some have suggested that Rosetta includes automatic translation algorithms from the one to the other. (Perhaps it's Xcode that they were thinking of. That would make more sense.)
And I have to reiterate something that I'm starting to believe more strongly: in terms of real-world performance, Mac buyers in the next twelve months are going to see just as smooth a ramp in speed as they've always been accustomed to seeing. Between now and the switchover from top-end G5s and G4 PowerBooks to the Intel platform, said G5s and PowerBooks will continue to become faster, with probably one or even two speed-bumps between now and next June. And then, when the Intel boxes come out, they're not going to be hugely faster than their immediate predecessors from across the CPU divide: they'll have more raw oomph, yes, but that oomph will be mitigated by the lackluster performance of Rosetta-translated apps, which will account for virtually all of what customers have to work with until a) the app developers release new Universal Binaries, and b) the customers buy them. In fact, right out of the gate I wouldn't be surprised if the first Intel boxes are markedly slower in most everyday tasks than the G5s sold the previous day. But that would change with time, on the same hardware, as the apps made the transition.
The saving grace, perhaps, is that Apple controls so much of the core user experience—iTunes, iPhoto, iWork, Mac OS X itself—that it can guarantee that Intel buyers out of the box will be using Rosetta only for third-party stuff. And since most of the big-name third-party vendors will have made the transition to universal binaries by I-Day, there will at least be a clear upgrade path available for Office and Photoshop users, and though Rosetta will be no powerhouse, at least it will only come into play a minority of the time.
And there definitely seems to be more to that bear hug by Paul Otellini than mere showmanship. Chris M. writes:
Intel has some really, really smart engineers who are sick of hearing
how great Power PC is, and are tired of being hobbled by the needs of
the Wintel manufacturers. Apple meanwhile saw that IBM just wasn't
responding to their needs either. I wonder if this whole alliance might
have started with a few engineers from Intel and Apple getting together
one night for beers....
Reported from many directions, including AnandTech, is this idea that Intel needs Apple as much as Apple needs Intel:
For the longest time, Intel has been promising usage models and concept PCs that we all wanted, but would never surface. PC vendors focused on producing the cheapest system possible, while dealing with backwards compatibility and standards compliance with a huge install base - effectively, making change difficult. Look at how long it has taken us to transition away from the Parallel and Serial ports on PC motherboards or the move to SATA drives. With Apple, Intel finally has a partner that is willing to adapt to change at a much more rapid pace and one that can implement new technologies extremely quickly, thanks to a small, agile user base.
And via Mitchell M., Cringely makes an odd connection that I haven't seen elsewhere, but that would explain quite a bit (including Apple's blackout of AMD during the keynote):
The crowd this week in San Francisco at Apple's World Wide Developers Conference seemed mildly excited by the prospect of its favorite computer company turning to Intel processors. The CEO of Adobe asked why it had taken Apple so long to make the switch? Analysts on Wall Street were generally positive, with a couple exceptions. WHAT THE HECK IS GOING ON HERE!? Are these people drunk on Flav-r-Ade? Yes. It is the legendary Steve Jobs Reality Distortion Field at work. And this time, what's behind the announcement is so baffling and staggering that it isn't surprising that nobody has yet figured it out until now.
Apple and Intel are merging.
Think so? I'm sure weirder things with fewer telltale signs in the marketplace have happened, but I'm at a loss to think of any. More:
Microsoft comes into this because Intel hates Microsoft. It hasn't always been that way, but in recent years Microsoft has abused its relationship with Intel and used AMD as a cudgel against Intel. Even worse, from Intel's standpoint Microsoft doesn't work hard enough to challenge its hardware. For Intel to keep growing, people have to replace their PCs more often and Microsoft's bloatware strategy just isn't making that happen, especially if they keep delaying Longhorn.
Enter Apple. This isn't a story about Intel gaining another three percent market share at the expense of IBM, it is about Intel taking back control of the desktop from Microsoft.
Intel is fed up with Microsoft. Microsoft has no innovation that drives what Intel must have, which is a use for more processing power. And when they did have one with the Xbox, they went elsewhere.
I had wondered whether the last straw for Steve was seeing all three game console companies suddenly announce new products based around PPC970-hybrid chips running at higher speeds than the fastest Power Mac available today; whether the 3.2 GHz vs. 2.7 GHz advantage comes from in-order execution or not, Steve may have decided that if IBM was that interested in throwing its weight into the embedded market rather than plunking down for a single-customer low-power laptop-bound G5, just as Motorola had always been more interested in cellphones than in developing its own G5, then now was the time to jump (rather than lead us through another sick period of fleeting promises like the G4 days). But what if Intel was as interested in breaking free of Microsoft as Apple was of securing its processor future? What if this deal was Intel's idea?
What with those huge leaps in Mac sales that Steve pointed out in the keynote over the last two months (while PC sales have remained flat), and what with iPod profits making the company feel its oats, it's certainly in a seldom-precedented position of strength if it's going to make such a risky move, and one that's so clearly going to result in such sucky sales for the coming year. But if Intel had been the one making overtures to Apple, putting up an offer Steve couldn't refuse—maybe that explains how he was able to keep the quaver out of his voice when he stood up there on that stage. Not to mention why his smile was as self-assured as ever.
UPDATE: Can this inevitability possibly be intentional? This from the same company that released iTunes 4.0 with a "Connect to Shared Music Library by IP" feature and apparently didn't think people would have incorporated stream-hijacking utilities into it within twelve to fifteen nanoseconds?
Unless they've learned more about Machiavellian espionage techniques in the last year than the CIA appears to have forgotten since 1970, my money's on "no". Just a happy accident, if this ends up only having positive repercussions.
UPDATE: AccelerateYourMac (via Bill B.) has some detailed, hands-on performance assessments of the Intel-based developer-kit Macs, giving us some idea of what we might expect to see native Intel apps like iTunes and iMovie doing a year from now (hint: they'll rock out). Also some more insights into what Rosetta does and to what extent. The tests they've been running "peg it at between a dual 800 MHz G4 and and a dual 2 G5 depending on what you are doing". And apparently "There are a lot of apps from big names running in the Compatibility lab already". Just imagine that atmosphere: geeks with company-logo knit shirts are carrying around their companies' flagship products in source-code form on iPods and LaCie drives; they're taking them into labs underground at Moscone, and before the week is out there are compiled binaries running on a new computer platform that nobody knew existed a week ago. That's what Mac geekery is all about: it's not just people explaining why one-button mice are superior or smugly spelling "Microsoft" with a dollar sign; it's about movers and shakers creating world-beating technology while at the same time being true believers. Sorta reminds me of Gothic cathedrals or the space race.
This isn't without its downsides, though: reportedly, the Intel Macs—at least the current dev kits—have a Phoenix BIOS instead of Open Firmware. Groan, shudder. Open Firmware is evidently gone forever, but this is an ugly, ugly compromise, and probably the worst concrete example I've seen yet of the lurches back to the stone age that we feared since the keynote confirmed what at the time was a nightmare fantasy. There's talk of a new Intel initiative for an Open Firmware-like replacement (EFI)—maybe this is one of those things Intel has been trying to do for years, but that Microsoft never let them realize their ambitions—but as things stand now, those boxes are pretty uninspiring to people weaned on boxes with "programmer's buttons" and complete device trees at POST.
Oh yes—and whither "sleep" mode for laptops? I imagine Intel's mobile chips can do all the bang-zoom stuff that G3/G4 laptops can do when you close the lid, and if not, perhaps that's another thing Intel wants freer rein to work on...
UPDATE: Jeff Harrell weighs in still further, now that all the breathless speculation of Intel takeovers and Microsoft-territory smash-and-grabs and Steve Jobs vacationing on a Pixar-rendered beach planet have settled into our stomach linings. It's definitely looking like nothing more sexy than a somewhat ramped-up bid for the same consumer market served by the Mac mini; but that's no small deal in the grand scheme of things.