Chip tests: Itanium wins some, loses some

Two key chip speed measurements using standards set by testing organization Standard Performance Evaluation Corporation (SPEC) show the 800MHz Itanium in some cases neck-in-neck with Sun Microsystems' newest UltraSparc III chip and in some cases soundly beating it. Sun is Intel's chief competitor when it comes to trying to win a place for Itanium, a chip Intel announced Tuesday.

One measurement of "floating-point" mathematical calculations shows Itanium soundly beating UltraSparc III--even the delayed 900MHz version due to arrive shortly. But when it comes to more ordinary "integer" operations, Itanium is behind Sun's chip and even Intel's own Xeon server chip.

"The SpecFP (floating-point) number was just phenomenal, but the SpecInt (integer) number was very weak," said MicroDesign Resources analyst Kevin Krewell. "Intel has not been able to provide a very good reason why that number was so weak."

"We feel (integer performance) is competitive today with UltraSparc III, and we expect it to get better with McKinley," the second-generation Itanium model due in the first half of 2002, said Intel spokesman Seth Walker, who said Itanium shines on higher-level benchmarks such as database performance or data encryption.

Itanium is the first of a new high-end line of chips from Intel that speak a totally different language from the existing low-end Celeron, mid-range Pentium and high-end Xeon lines. The existing products are 32-bit chips, but the Itanium is a 64-bit model, the key difference between the two being the ability to accommodate vast amounts of memory.

Itanium, delayed several times from its planned arrival in the 1990s, was spawned by a collaboration between Hewlett-Packard and Intel. The chip is a key element of Intel's strategy to bring to top-end servers and workstations its philosophy of high-volume manufacturing and support from multiple computer makers. The Santa Clara, Calif.-based company has been largely locked out of the market, which is dominated by 64-bit chips from Sun, IBM, Compaq, Hewlett-Packard and SGI.

Piper Jaffray analyst Ashok Kumar believes Intel ultimately will succeed.

"Itanium itself is a solid product, but it doesn't blow the doors off the competition, as Intel had initially hoped (and hyped)," Kumar wrote in a report Wednesday. "But it is best viewed as the first step in a long line of processors that will eventually dominate the high end of computing, pushing aside current competitors such as IBM's Power family and Sun's Sparc processors."

Itanium beats even Compaq's lauded Alpha chip, Terry Shannon, author of the Shannon Knows Compaq newsletter, said in a note Tuesday. But Shannon predicted Alphas running at higher clock speeds--1GHz chips are expected by mid-July--"will restore Alpha's performance leadership across the board." Itanium didn't convince Shannon that its design "will doom rival architectures to near-term obsolescence."

Krewell predicted IBM's coming Power4 chip, due to arrive in servers this fall, also will keep the pressure on Itanium.

But delays have hampered Itanium's prospects. "If this product had come out last year, it would have been a much more impressive part," Krewell said.

More than just the numbers
But raw performance is just one of a host of factors Itanium faces in the marketplace. Of greater concern has been the fact that software must be completely rewritten to take advantage of Itanium's performance. Although older software designed for the earlier 32-bit chips will run, Itanium performance on such tasks is dismal.

The situation hasn't escaped Sun's notice. "We think that this announcement actually enhances our competitiveness because it highlights the transitions that customers and vendors need to go through with the Itanium," said said Nancy Weintraub, director of marketing for system products at Sun. Customers must recompile their software for the new chip and re-certify it for the new systems, she said, whereas Sun customers are guaranteed that software written for earlier UltraSparc chips will run unmodified on the UltraSparc III.

Intel and its business partners have only just begun the laborious process of getting software ready for Itanium. The first version of Microsoft's Windows XP for Itanium workstations isn't expected until Oct. 25, and higher-level software such as databases likely will arrive even later.

"We do not expect Itanium to be broadly deployed," Kumar said. Though some scientific users will buy the systems, "most businesses will purchase only a few Itanium systems for software development and testing."

Intel is girding for a struggle with Sun and others that could go on for years, though.

Broader use will come with the second-generation McKinley chip due in systems a year from now. And another successor, code-named Madison and due in early 2003, is expected to be three times faster than the first Itanium, Kumar said.

Though introducing a totally new design is very hard on programmers and customers who must totally rewrite software, Sun and others will see their designs run out of steam and likely will have to take the drastic step Intel took when moving to the Itanium, Krewell said.

"At some point in time, you've got to break the previous architecture and start from scratch," Krewell said. "I think the UltraSparc architecture is getting old."

Intel has pushed its comparatively ancient 32-bit designs remarkably far, though, and Weintraub defended Sun's design. "We feel very strongly about the longevity of the Sparc chip," Weintraub said.