If Itanium fails and Hammer succeeds, Intel is in a world of hurt.
"At the AMD tech webcast a couple of months ago, near the end, an analyst asked Jerry what was next, what's highest priority, etc., and he yelled out Hammer, Hammer, Hammer. He mentioned nothing else. That was it. I repeat, Hammer is essential to the survival of AMD as we know it, this year-next year. Itanium, on the other hand, is not absolutely essential to the immediate future of Intel. Hammer: heart and soul. Itanium: not just a nice to have, but Intel doesn't die without it. It's a matter of overwhelmingly important products to a company. Hammer is, Itanium isn't. "
If Itanium (including McKinley, of course) fails to the point of losing the support of HP, IBM, and others, and if AMD pulls of the "stopgap" Hammer (an Athlon with more 64-bit enhancements), then will be very, very bad news for Intel.
This will mean the mantle of leadership will have passed to AMD and its partners (Via?), with Intel forced into either licensing Hammer or doing its own "stop gap" product. (Either the Register r the Inquirer or both had something about Intel quietly starting a stop gap project...don't know if this is true.)
We saw this with the 8800/432 project around 1977. A stop gap project was hastily put together to produce a product to fill the niche that the 8800/432 was apparently going to be too little/too late to fill. Steve Morse was given about 3 weeks to spec the part that became the 8086.
Until a few weeks ago, I assumed that the lack of any obvious stop gap projects for the past few years meant that internal studies--and customer reactions--had made such a stop gap unnecessary. After all, the logic goes, HP, IBM, Dell, etc. have had a long, long time to evaluate the Itanium architecture.
However, several things are now making me worry quite a bit:
1. Dell's dropping of the IA-64 workstation product, apparently due to stillborn sales. This by itself is not conclusive, as Dell is a "high volume" company, not a big iron company, and the decision to drop a particular box is not terribly significant. For example, they may be regrouping with a McKinley version. Or they may wait for volumes in general to increase.
2. IBM continues to push their Power4 architecture for high-end use in their own systems. Why is the IA-64 not gaining in such uses?
3. According to someone whose expertise I value, who works on the IA-64, "it's just a dog." (I can't say much more for fear of making his identity apparent to some Intel employees who may be reading this.)
I hope he's wrong. But despite some reports of good "big iron" performance, for things like Web commerce key generation and verification and Oracle-type apps, it doesn't look like the VLIW/EPIC ISA is offering a leap forward in performance.
If a 2.2 GHz P4 or 1.7 GHz Athlon outperforms a massively-large chip like the Itanium, or even the McKinley, why bother switching to a new ISA unless the 64-bit addressing is critically needed?
4. The chaos underway in the HP/Compaq deal. These were two of the strongest names behind the IA-64, both from the HP role as co-designer and heir to PA-RISC and from the Compaq/DEC role as heir to Alpha and legacy (VAX) OSes. If the deal falters, can Compaq go back to the Alpha? Probably not. But it may hurt the IA-64.
On the positive side, Intel has lined up a whole lot more software support and chip support for IA-64 than AMD has for Hammer. Hammer has not even taped out yet, putting it at least several year behind Itanium. And while we have ports of major "big iron" and workstation products for IA-64, we have only promises of support for Hammer. No doubt vendors are looking to see if IA-64 stumbles and Hammer works within the next six months.
So I'm not predicting the failure of IA-64. Intel has committed a vast amount of resources to it, as has HP. And the very points I'm making here, about what Intel would stand to lose if IA-64 were to fail and Hammer were to be seen as the stop gap nextgen (pun intended) architecture, well, these points must be uppermost in Intel's mind. Which means they'll probably do what it takes to ensure that any performance problems with the IA-64 implementations are fixed muy pronto.
Still, I worry. And I don't agree with the point that "Itanium, on the other hand, is not absolutely essential to the immediate future of Intel. Hammer: heart and soul. Itanium: not just a nice to have, but Intel doesn't die without it."
While Intel will not die without Itanium, it will be in a very bad position indeed. As a longterm holder of Intel stock, I hope they are giving this their highest attention.
--Tim May |
