Intel, HP Fund Effort To Boost Itanium Compiler Performance

In an effort that could boost the prospects for Intel's high-end Itanium processor, Intel and Hewlett-Packard are funding an effort to develop improved compilers capable of generating faster-running code for the chip.

The two companies are working under the umbrella of the Gelato Federation, a consortium of supercomputer centers and research organizations intent on using open-source software to field large clusters of 64-bit systems built around Itanium.

"There are two big things that will determine Itanium's future—price and performance," said Wen-mei Hwu, the compiler expert and electrical engineering professor at the University of Illinois at Urbana-Champaign who's a leader of the compiler effort. "What Gelato is doing is trying to raise the performance part of the equation and bring Itanium to a wider audience. The members of Gelato are all performance-critical users of Itanium."

Specifically, work will proceed on improving two software tools: the OpenImpact compiler developed by Hwu at the University of Illinois specifically to run on Itanium, and the Free Software Foundation's more general GNU compiler collection. Commonly known as GCC, it's a multi-platform set of compilers for C, C++, Fortran and other languages.

"We're working with Intel and HP to improve the performance of GCC on Itanium," explained Hwu. "The GCC compiler has a very wide base. If you want to reach the widest possible audience, GCC is the way to go. My project is to improve the user-visible compiler technologies, whether it's improving Linux performance or applications performance."

The first concrete milestone of the effort will be timed for May, when the Gelato Strategy Council is scheduled to meet in Cupertino, Calif. "The goal is to demonstrate that we can improve the performance of the 'Vanilla' [test] suite of open-source critical applications by then," Hwu said.

Whether the ability to generate faster code will enable Itanium to regain the limelight is anyone's guess. Intel argues that the chip has never left the spotlight. Intel says that more than 100,000 Itanium units were sold in 2003, the last year for which it has provided figures. "We will not provide sales numbers for 2004 and beyond," said Intel spokeswoman Erica Fields, citing the fact that Intel doesn't break out individual chip sales. "But sales of Itanium-based systems totaled $1.4 billion in 2004. It's doing great. Eighty-three of the top 500 supercomputers are based on Itanium [and] more than 40 percent of the top 100 global companies have deployed Itanium."

Nevertheless, the 10-year-old architecture, which was developed jointly by Intel and HP, has sometimes been publicly outshone by IBM's competing Power processor. IBM beat Intel to the punch on the dual-core front, rolling out a two-CPU Power 4 chip in 2001 and a dual-core Power 5 in March 2004. Intel plans to field its first dual-core Itanium 2, code-named Montecito, later this year. The part will ship in volume in 2006, Fields said.

Other experts wonder whether Itanium will remain a high-end processor, or whether it can achieve broader penetration. "The biggest issue is really whether Itanium is going to be a mass-market server processor," said Hwu.

A major factor in that regard is price. Although Itanium-based servers can be bought for as low as $1,300, many popular offerings are available at system prices in the $3,000 and $10,000 price ranges. (The Itanium processor isn't commonly sold to end-users, so processor pricing isn't considered a relevant metric.) Intel declined to comment on possible pricing of the upcoming Montecito part.

Another element in the picture is Intel's plan to take Itanium and move it to a common platform with its mainstream Xeon server processor by 2007. Intel will bring the two lines together by creating a unified 64-bit motherboard with a new, one-size-fits-all chip socket. The move has a potential upside for Itanium in that it could enable Intel to move Itanium down-market and reposition it as a mainstream replacement for Xeon. In that scenario, customers would opt for Itanium instead of Xeon because the extra cost of upgrading would be insignificant.

However, there's also the possibility that the lower-priced Xeon could begin to move upstream and nudge Itanium out of the higher-end areas it currently calls its own.