Rupert Goodwins' Diary
Wednesday 23/11/2005
No problem whatsoever today in pointing you at a story in The Register, where Ashlee Vance has been pounding the carpet at the SuperComputing 2005 show. Lots of good stuff, but the star of the show was the Opteron. Then came the Itanium, then the Xeon. Not that anyone's talking much about the Xeon.
This is really bad news for Intel, which remains so keen to assure us that everyone loves the Itanium that any remotely negative comment from us guarantees a patient missive explaining that sales are up, excitement is building and it's only us naysayers in the press who refuse to believe it.
I can't speak for anyone else in the business, but I'd much rather have Itanium do well than otherwise. I remain a geek at heart and nothing pleases me more than to see something different take on a vigorous life of its own. There are too many x86 processors in the world — please don't let it be that forever. What's not to love with a fabulous new technology doing all sorts of clever things?
The market, that's what. It's clear that in terms of cost-effective performance, you get more for your money with lots of relatively cheap x86s than with fewer more expensive Itaniums — provided the rest of the architecture is in place, can handle fast memory and IO, and has good software support. On all these counts, Opteron wins — not only over the Itanium but the Xeon.
Exotica — and here one must include the Itanium — is a very high risk strategy for number crunching. Seymour Cray bet the farm on gallium arsenide being so much faster than silicon that it would give him an unbeatable advantage in supercomputer technologies, despite the fact it cost a great deal more to manufacture. The farm was toast. Silicon wasn't as good, but it was good enough.
The processor market is powered entirely by size: if you can make a million of something where your competition is making a thousand, the advantages are all yours. Without a mass market for the Itanium architecture, its stately prow will disappear under a sea of cheap Opterons. Supercomputer 05 is merely the latest high water mark.