Intel's Third "Bet the Company" Action
Saturn V writes a good analysis of today's Itanium launch:
"WHITHER ITANIUM ?
"After several years and billions of dollars Itanium has begun to ship. This is the biggest gamble made by Intel. If it succeeds it will do to the high end, big iron mainframe space, what the x86 did to the low end and mid region computing. Another analogy could be the first IBM 360. It received a lot of criticism for being late, but laid the solid foundation for the IBM hegemony."
Indeed, this is probably Intel's 3rd real "bet the company" action, the first one being the whole founding and silicon gate/MOS approach, the second being the decision to focus on microprocessors and become a world-class manufacturing company.
(The iAPX 432/BiiN was a larger expenditure, proportionate to Intel's earnings. "A billion invested in nothing," the nickname wags applied to BiiN, was a big deal in the 1980-1987 period, but the loss could be swallowed without derailing other programs. A collapse of the Itanium, should it happen, would leave Intel gasping for a comeback strategy.)
The temptation is always strong to just keep on with the same instruction set and "milk it" for as long as possible. (To their credit, DEC did not do this with the VAX, concluding it would not scale adequately. Hence the Alpha. A nice architecture, by all accounts, but the end result was that DEC was in such bad shape financially that a company formed several years after the VAX was unveiled was eventually able to buy them for relatively small change.)
This tension between "let's keep on doing what we know how to do" and "we'd better move in a new direction before it's too late" is an area that has had little good analysis.
"Today the hardware performance looks good, but is not compelling. For comparison the Digital Alpha in its hey day had excellent performance, but that did not prevent the slide of Digital into oblivion. But there are important differences, and Intel does indeed have a high chance of success."
At some point Intel had to jump to a new microarchitecture and cast off the "reptilian brain" remnants of the 8008. It tried 20 years ago with the 432, and failed. Should it have tried five years ago? Should it wait another five years? Tough questions, ones that are "above my pay grade."
"A. PERFORMANCE: The follow on product McKinley looks a lot better. The longer pipelines improves the clock rate, and the on chip cache improves performance even further. And the shrinks , Madison Deerfield etc, enhance the performance even more. But the important issue is how will the performance of the successors of Merced compare with the contemporaneous competing hardware platforms i.e. the follow on to UltraSparc, Alpha, etc. Developing a new processor chip is getting to be expensive. Once the Itanium makes a few inroads, a lot of competing vendors may not be able to afford the cost of staying on the performance treadmill, and may drop out of the processor business."
I believe this is already an established trend. MIPS is effectively gone as an architecture (SGI is a struggling Intel-based vendor now). SPARC has not become the mainstream processor once projected (remember the vision of it being used for more than just Sun machines). And PowerPC is mostly (90% plus) the Macintosh, with a few IBM workstations and parallel computers thrown in. Alpha is also ghettoized into the DEC/Compaq market, and even Compaq hedges its bets in a major way with Intel processors.
The economies of scale make it unlikely that any of these processors are yielding as high (normalized for some measure of complexity, naturally) as Intel and AMD processors are. (This was the second bet Intel made, in my view, that it could supply processors as a mostly processor company. A lot of us, as you know, didn't think Intel could be this successful in processors without having an ultra-high-volume "process driver.")
"B. COST: The Itanium platform promises the economy of scale a la x86. Everyone and their brother is building an Itanium platform ! Once it gets cheap, the virtuous cycle begins by increasing the market share giving rise to even greater economy of scale and thus increasing market share even further.."
Well, yes, everyone and their brother is building an Itanium box. However, this in itself does not guarantee success. (For example, in the early days of the PC, "everyone and their brother" included the Otrona Attache, Eagle Computer, Hyperion, and others. Lot of good these "everyones" did.)
Many of the Itanium boxes seem to be hedging of bets. After all, the chip is from Intel/HP, the OS is Linux or Windows XP (or in some cases some other flavor of Unix). It isn't terribly risky for Dell or Compaq to build a box: if the Itanium fizzles, they go back to P4s or Alphas or whatever. IBM still has their PPCs, Compaq their Alphas, and so on.
H-P seems to be in the most exposed position. A fizzling of the Itanium (Merced _and_ McKinley) would leave them scrambling to extend the PA-RISC line. (Apologies if I've gotten the name of their successor to PA-RISC wrong...they probably have some names I haven't been keeping up with.)
[stuff elided because my reply is already too long]
"Intel appears to have done the right things so far. The hype was necessary to get the Software Developers attention, but has also caused greater embarrassment about its publicized delays. Well , the IBM 360 had similar problems, but still achieved its objectives. We will all have to wait and see if Intel's gamble will indeed pay off."
Jumping to a new architecture (instruction set, more precisely) is always very risky. DEC did it with the Alpha, and ended up being a subsidiary of Compaq. ntel tried it with the 432 and failed. They had to fall back on various stopgaps for the creaky x86 and finally regained their solid footing with the 386 and its successors. The Pentium was just the culmination of their victory.
A big gamble. If the Itanium fails, the scramble to accelerate work on the Pentium 5, I guess.
If they succeed, Sparcs and Alphas and PowerPCs probably go away for good.
--Tim May |
