Intel's running out of Itanium options

Suppose you threw a party and no-one came? Intel has put the streamers out, the sandwiches have had the crusts cut off and the punch bowl is full. Yet the hall is looking bare. The Itanium party can't be considered a roaring success -- even one of the people who helped organise it has left. And now another friend has said "sorry, can't make it this time."

It's not been a good year. When HP announced that it was discontinuing its line of Itanium based workstations, things looked bleak. HP had been Intel's partner in developing the new architecture -- if they didn't think it was worth persevering with, who would? Intel has also had to concede that most people simply don't want or need Itanium-based servers at the enterprise level, instead launching the Xeon processor with Extended Memory 64 Technology, a 64-bit bodge onto a 32-bit processor. That leaves the big iron -- the last place that Itanium makes sense.

So what could hurt more than Microsoft saying that its Windows Server 2003 Compute Cluster Edition won't run on Itanium 2? The kind of computing that needs clustering is the kind of computing that needs fast, 64-bit processors, precisely where Intel is pitching the Itanium. Microsoft palpably disagrees: while its clustering software may be looking at so few sales that any form of extra development effort is unwelcome, the fact is it doesn't see Itanium as the most important chip for big installations. That runs exactly counter to Intel's message, and that's bad news.

It's not as if 64-bit chips aren't selling elsewhere in the market. AMD's Opteron and Athlon 64 processors are selling well, and Sun and IBM continue to produce their own 64-bit chips. Itanium can't compete with these other chips: Sun and IBM both have in-house operating systems that are closely tied into their hardware and are aiming at a market where most of the software running is written for a specific task. AMD's 64-bit chips run 32-bit code natively, thus ensuring backward compatibility but with a performance boost. More importantly, AMD's chips are cheaper. In bang-per-buck terms, Itanium just isn't cutting the mustard: in providing an upwardly compatible migration path, it's not even on the farm.

If Intel can't make and sell the Itanium family for less, it needs to rethink its plans for 64-bit. Few people can justify spending more on a processor that runs their existing applications slower. AMD will take the mainstream market, while the high-end will be snapped up by companies that produce the entire machine -- not just one component. We may all be heading for 64-bit computing, but it won't be by buying Itanium. That leaves Intel with a hard choice: abandoning the architecture would be massively embarrassing, bleeding to death would be worse.