Analysis The marketing theme of IDF this week has been customer choice, but vendor self-interest would be a more accurate description of the affair.
Intel muddled its way through the birth of the new Xeon (now enhanced) processors. CEO Craig Barrett, during a IDF keynote, slipped the 64-bit extensions on a slide covered with future Xeon improvements. The 64-bit chips were given the same billing as a new front side bus. Server chief Mike Fister* also tried to apply his aw shucks approach to the technology, using words such as "natural extension" and "complementary." He glossed over the very real challenges Xeon (now enhanced) poses to Intel's struggling Itanium line, how Intel will for the first time deal with serious competition from AMD on the server or why customers should not have known about this sooner.
Intel keeps chanting that the software "ecosystem" is far too young for x86-64-bit computing just yet. And, while the absence of even a 64-bit Microsoft OS makes that clear, you have to weigh the stance against what Intel's own OEMs claim. IBM and HP insist they have seen a sudden spike in the need for 64-bit Xeons and that the processors give customers more options in the short and long term.
"This puts choice in the hands of our customers . . . This is a classic example of how industry standard servers respond to customer demand," said Don Jenkins, who leads HP's high-end server marketing, during a speech here.
"We think the 64-bit extensions help in protecting customers' investments," said Susan Whitney, general manager of IBM's xSeries server group.
IBM can make a pretty strong case for understanding what's going on in the x86-64-bit market. It apparently picked up on the rise in customer demand a bit earlier than Intel and HP when it began offering servers based on AMD's Opteron chip when the product first launched last year. (Something Intel was not pleased about.)
IBM has seen most customers pick up Opteron for 32-bit software. This lets customers make use of Opteron's strong performance now and opens the way to move to 64-bits, if this becomes a priority.
But didn't Intel and HP customers deserve to know that this 32-bit to 64-bit path would be an option?
Yes, the rumors of Yamhill have floated about for years. Anyone keeping an intense eye on the chip scene figured the Xeon extensions were coming. Most, however, pegged the technology's release at sometime in 2005 - not midyear 2004.
Intel and HP allowed the speculation to run its course, buying time for Itanium.
On one level, this claim doesn't make much sense, and we are happy to admit it. HP's PA-RISC and Alpha customers are not likely to move their software onto x86 servers. HP has paid too much attention to the EPIC porting and tuning process for that to happen. So confirmation of Xeon extensions would not make too much difference to these users.
But, in the end, HP's Unix customers are only part of the Itanium equation. They are the ones being forced - on HP's timeline - to shift servers. It's the Windows and Linux crowd that have far more options.
HP has been crowing for years that it can run Windows, Linux and Unix on the same Itanium box. HP even says that one-third of its Itanium server shipments have gone to Windows customers.
You, however, have to wonder if all this Itanium attention would be paid to Windows and Linux, if HP had confessed to the Xeon extensions a bit earlier. You have to wonder what would have happened to shipments of the Madison version of Itanium 2 if HP's vast customer base knew what was coming and when.
It seems fairly obvious that Intel and HP allowed AMD to take the lead in x86-64-bit computing in order to protect Madison for as long as possible. After all, only 100,000 Madison processors shipped last year - almost all of those flowing through HP. This is up from a total of 10,000 units of the first two Itanium iterations. Intel and HP could not afford to see Madison sink like its predecessors and needed to claim there was no other "real" 64-bit option.
How else can you explain HP's sudden embrace of x86-64-bit systems? HP swung from total denial to picking up both Xeon (now enhanced) and Opteron in one month. Again, it says this is a response to heavy customer demand - the type of demand that apparently did not exist in January.
In the end, Intel and HP played their customers for suckers. They held off on extensions for as long as they thought possible. But with IBM and Sun Microsystems starting to enjoy a little too much attention, the vendors figured they needed to act.
Intel and HP gave their customers a decade to decide whether or not Itanium was the right choice, but only awarded 3 months of lead time to the Xeon crowd.
We can only wonder how the army of ISVs performing EPIC ports in a down economy feel now. Resources might have been better spent on prepping code for the vast Xeon (now enhanced) market as opposed to the wobbly Itanic one. ®