Intel’s Itanium: Plan B in the works

The company is looking at building capabilities into the next version of the Pentium family, code-named Prescott, that will allow the chip to run 32-bit applications, the same sort of applications that currently run on Windows machines, as well as 64-bit versions of those applications, according to a report in The San Jose Mercury News. Sixty-four bits is the amount of data the chip can process during one clock cycle. Overall, 64-bit chips can greatly improve performance and are necessary for applications such as data mining.

It's unclear if such a 32-64 bit chip would ever hit the market, or even if it's technically feasible to add such features into Prescott, according to analysts. Intel engineers, though, are tinkering with the concept, according to the Mercury News and Ashok Kumar, an analyst at U.S. Bancorp Piper Jaffray.

An Intel representative declined to comment on unannounced products but said that the company in general maintains backup plans.

"There are any number of plan Bs," said a representative, adding, "Itanium is the basis of an architecture that will span more than two decades."

Arcane as it sounds, the release of a chip with those capabilities would be a significant event in the server world.

Intel currently promotes its Itanium processor for servers that run 64-bit applications. These servers are supposed to compete against similar 64-bit machines from Sun Microsystems. Delays, uninspiring performance, and a lack of applications, however, have combined to make Itanium sales anemic.

A Pentium chip with so-called 64-bit extensions would be compatible with the existing Windows server software, which would ease commercial adoption. Such a chip would also stem interest in Hammer, the code name for a 32-64 bit chip from Advanced Micro Devices. AMD Chief Executive Jerry Sanders has called such a chip from Intel "my biggest fear."

Since last November, computer executives and analysts have said Intel would inevitably try to develop such a chip. Some computer companies have even actively encouraged Intel to put greater emphasis on a 32-64 chip. The existence or status of any projects has continued to be vague.

Of course, such a chip would face an uphill climb in getting to the market, according to analysts. Intel has spent massive amounts of time, energy and money building the Itanium. Suddenly offering an alternative chip that would be more widely compatible with the world's software would drastically undercut those efforts.

"It makes a lot of sense for them to do it, but there is zero incentive," said Dean McCarron, principal analyst at Mercury Research. "It could happen if AMD starts getting credible positioning in the entry-level server space with Hammer."

Technically, integrating such capabilities into Prescott, due out in late 2003 or early 2004, would not be easy either. "I'd be real surprised if they could drop it in without a major overhaul," said Nathan Brookwood, an analyst at Insight 64.

This chip would also require both hardware makers and software developers to come up with products that would be able to take advantage of the 32-64 bit design. Some of these companies, including IBM and Hewlett-Packard, have already invested substantial amounts in Itanium projects.

Despite the obstacles, the chip would have at least one strong argument in its favor: Interest in Itanium remains weak. The first version of the chip debuted in May of last year. In the third quarter of 2001, only 500 servers containing Itanium were shipped, according to IDC.

McKinley, a new version of the chip, is currently being launched on a test basis and will come out commercially in the middle of 2002. McKinley is expected to perform far better than the current versions of Itanium.

Questionable demand
Nonetheless, demand remains fairly uncertain. Compaq Computer executives, for instance, have said that companies will spend most of 2002 trying to popularize Itanium, with volume sales beginning in 2003 and 2004, when the successors to McKinley, called Madison and Deerfield, arrive.

ServerWorks, which manufactures chipsets for Intel servers, said it likely won't come out with Itanium-compatible parts until Deerfield. "We're just waiting for volume to materialize," said Kimball Brown, vice president of business development at ServerWorks.

A major stumbling block for Itanium has been a lack of applications. The chip is built around an entirely new architecture, which goes by the acronym EPIC, that requires brand-new software. Itanium chips can actually run standard Windows programs; the chip includes an integrated Pentium chip. Performance, however, is not very good, according to most analysts and computer experts.

Although many new Itanium applications are expected to arrive with McKinley, the software available will be dwarfed by the number of servers with standard Intel chips. Software developers also don't have the dollars to invest in Itanium programs right now.

Customer interest for Itanium servers is "effectively zero," Joe Marengi, senior vice president of Dell Americas, said in an interview in November at Comdex. "The investment involved in the transition is huge." At the time, Marengi suggested that one could argue the relative merits of a hypothetical 32-64 bit chip.

Few companies have ever successfully switched from one architecture to another. Intel, in fact, attempted to switch from its "x86" architecture, the architecture behind its microprocessors, in the early '80s. Many of the company's best designers were put on the Sierra project, a chip based around a budding new architecture that would succeed the 286 chip, according to Brookwood. In the end, however, the company went with its plan B, an extension of the 286 called the 386. Pat Gelsinger, one of the leaders on the 386 project, has since become Intel's CTO.

Ironically, Intel contemplated doing a 32-64 bit chip in the early '90s, according to former company executives, but decided to concentrate on chips based around the then-emerging EPIC architecture.