senior VP, Intel
The chipmaker has been roundly criticized for that incompatibility, which hampers software support, but Intel's strategy is sound because Itanium serves a market largely unaffected by the x86 expansion, said, the senior vice president in charge of Intel's digital enterprise group.
Demand for servers using x86 processors has grown dramatically in recent years, but that growth has largely tapered off, leaving the market for powerful "big iron" machines that use Intel's newer Itanium chips unthreatened, he argued at thehere last week.
"We're projecting a very robust, large market for the biggest iron, for a decade-plus to come," Gelsinger said in an interview. "That's where Itanium is aiming."
It's a reasonable approach, said Gartner analyst Martin Reynolds. "Over the next few years, I think Itanium is going to start looking really interesting," he said.
It's no surprise that Intel is angling for influence in servers. The powerful machines are the brains of computer networks, used for everything from hosting Web sites to executing bank transactions.
Intel believes x86 chips such as Xeon, although increasingly powerful and widespread, won't penetrate the market for high-end servers for years to come.
Intel argues Itanium has a strong opportunity in the high-end market, despite acknowledged missteps in the past with the chip family. Many server makers aren't aligned with Intel's view of the market, though.
But Gelsinger acknowledged difficulties in the Itanium launch, which has been afflicted by delays, poor initial performance and software incompatibility with x86 chips. Intel has scaled back its earlier plans to make the chip as widespread in servers as Pentium is in desktops, and now bills the chip only for high-end machines.
"Obviously, the Itanium ramp is more difficult than we forecasted when we started," Gelsinger said.
And Intel might not have made the same decisions had it thought more carefully about the Itanium challenge, he said.
"Introducing a new architecture is very hard. Had we laid out the business plan and really saw how long it took to get there, management and the board of directors might have made different decisions," Gelsinger said.
The issue is put to rest now, though, he said. "We now made the investment, and the industry is adopting it. We're out of the woods. This dog hunts."
x86 in the driver's seat
Gartner's Reynolds believes that x86 servers have captured the attention and momentum of the server market. "The small servers have destroyed the growth opportunity for big iron," he said.
At the same time, Itanium has the potential to grab a bigger share of the market that remains, at the expense of Sun Microsystems' UltraSparc and IBM's Power, he said.
So far, Itanium machines have made a modest dent in the server market. Server customers purchased $1.6 billion worth of Itanium machines in 2004, up from $448 million in 2003, Reynolds said.
By comparison, x86 server purchases grew from $21.1 billion to $24.6 billion between 2003 and 2004. Sparc system purchases shrank slightly from $6.9 billion to $6.7 billion, while Power purchases shrank from $7.2 billion to $7.1 billion, Reynolds said.
Hewlett-Packard, which co-developed Itanium, has separate x86 and Itanium server families that are aligned with Intel's view of the market. But the three remaining major server manufacturers, which along with HP control 80 percent of the server market, have differences of opinion.
IBM, the largest server maker and the acknowledged leader in powerful machines, is pushing--in other words, big-iron x86 machines. Dell believes and instead believes groups of linked x86 machines will rule the roost. And Sun is working on from Intel rival Advanced Micro Devices.
IBM and Sun, of course, can't be expected to warmly embrace an Itanium processor aimed squarely at their core product lines.as effectively dead, while .
Intel doesn't hesitate now to call out. "We are going to compete head-on for every socket versus Power," Gelsinger said.
Of IBM's Itanium decision, he added, "We're disappointed, but it's not an irrational strategy," given that IBM has Power to promote. And the companies still cooperate when it comes to Xeon, the Pentium big brother designed for servers. "IBM's a huge customer for us for Xeon. It's their fastest-growing business line," he said.
Intel's goal is to provide a high-end processor for as many server makers as possible. Even with IBM's departure, Intel still has Itanium partnerships with Fujitsu, Hitachi, HP, NEC, Silicon Graphics Inc. and Unisys. "Our job is to give everybody an opportunity to compete in that high end," Gelsinger said. Those customers must ask themselves whether they want to use Itanium or to "live on IBM's intellectual property," he said.
In fairness, it's not quite that simple. Fujitsu, for example, also has a major Sparc processor development program in cooperation with Sun.
Different memory designs could help Itanium, Reynolds said. Today's Itanium requires vast amounts of high-speed cache memory to be competitive, but a retooled memory system could make the chip smaller and more competitive, he said.
Intel's biggest chip competitors--AMD, IBM and Sun--all have moved a crucial computer component, the memory controller, directly onto the central processor, which quickens memory access speeds. But it's likely that they won't have that advantage forever.
A probable time frame for making that change at Intel is 2007, when the chipmaker plans to unify the interface between Xeon and Itanium chips, a move that would make it easier for computer designers to make dual-use components.
Gelsinger wouldn't comment on when or whether Intel will include an on-board memory controller, but he did say, "Over time, you will see us make adjustments to memory hierarchy."