Intel's Secret Weapon!

[Joseph Zollo]

http://www0.mercurycenter.com/business/top/038013.htm

Tucked away in Hillsboro, Ore., a small team of Intel engineers has been quietly working on a chip technology that the giant semiconductor maker hopes will never see the light of day.

Intel's Yamhill Technology is a secret weapon against upcoming chips from rival Advanced Micro Devices. It's also a hedge against the possible failure of Intel's flashy new Itanium chips for computer servers, which have so far gotten a disappointing reception from customers and partners.

Unlike the Itanium, which uses a novel design that has some difficulty running software written for Intel's well-known Pentium family of chips, new chips with the Yamhill feature could easily handle the older programs as well as more sophisticated, memory-intensive games, database programs and scientific applications.

While having a Plan B might seem to be simple prudence -- especially at Intel, with its famed culture of paranoia -- the initiative is controversial within the company's Santa Clara headquarters.

Some Intel executives believe the Yamhill Technology's existence will be seen as a tacit admission that the Itanium, which took an estimated $1 billion and seven years to develop, might be a flop. In addition, the swashbuckling Intel doesn't like to admit that the much smaller AMD, based in Sunnyvale, could pose a real threat.

The Yamhill features are being built into the next version of Intel's Pentium chip, code-named Prescott, with an option to turn the features on or off. In 2003 or 2004, when the Prescott chip is expected to be available, Intel will evaluate AMD's offerings and the success of the Itanium and then decide whether to activate the Yamhill code.

But Intel's executives hope they never have to turn on the Yamhill features, said an engineer who worked on the secret project and has since left the company.

``I'd presume that they would only begrudgingly do anything with it at all. It will only be if the competition forces them,'' said the engineer, who agreed to discuss Yamhill only on condition of anonymity.

At stake are billions of dollars a year in sales of the high-speed chips that serve as the brains for computer servers, a market which Intel and AMD have both targeted for future growth as personal computer sales flatten.

With its upcoming Hammer family, AMD is betting that server buyers want 64-bit chips compatible with Intel's popular Pentiums and Xeons and AMD's Athlons. Those chips, used in most of today's PCs and low-end servers, operate at 32 bits and are known as x86 designs, a reference to the names of early PC microprocessors, which all ended in the digits ``86.''

Intel is wagering on the Itanium, which also processes 64 bits of data at a time and has the added ability to execute many instructions simultaneously. The Itanium is designed to handle huge computing problems that require large amounts of memory, but it runs today's software more slowly.

Yamhill, named after a small town near an Oregon river of the same name, is Intel's backup plan. The extra features, called extensions, will make the Prescott chip a 64-bit chip compatible with Intel's traditional x86 designs, like AMD's Hammer, in case that's what customers really want.

Intel officials won't publicly acknowledge the Yamhill project, despite industry rumors of its existence. In the beginning, the small group of Yamhill engineers at Intel's Ronler Acres campus in Hillsboro worked in such secrecy that they could only refer to the technology as YT in e-mail or conversation. Talking in the halls about YT was forbidden, and engineers could not even tell their colleagues what they were working on.

An Intel spokesman declined to comment for this article, saying the chip maker doesn't comment on unannounced products.

But Ashok Kumar, an analyst with U.S. Bancorp Piper Jaffray who closely follows the chip industry, said he knew some details of the backup strategy. ``The thinking is it would give them more elbow room in extending the current architecture,'' he said.

Kevin Krewell, a senior analyst at MicroDesign Resources in San Jose, said he fully believes that Intel has a ``skunkworks project'' to extend the current x86 family into the 64-bit computing world.

``They would be foolish not to have a fallback plan,'' he said, adding that he does not have any direct knowledge of such a project. But ``Intel cannot admit that it is going on because it would undercut their Itanium program.''

In 1993, Intel decided that as computer programs got bigger and more complex, its microprocessors would need to do things differently. In the past, it had based its entire x86 chip family on a design that used Complex Instruction Set Computer (CISC) principles, which process instructions one at a time.

Intel also wanted to expand beyond its slowing PC chip business and enter the world of big, high-end servers, dominated today by Reduced Instruction Set Computer (RISC) architecture, which is used by today's leading makers of big servers, including Sun Microsystems, IBM and Hewlett-Packard.

RISC chips not only process multiple instructions at the same time but also run at 64 bits, meaning they can simultaneously process twice as much data as the standard 32-bit Intel chips. The 64-bit chips are particularly well-suited to server programs that require huge amounts of memory, such as big databases or computer-aided design software.

To meet the server challenge, Intel partnered with HP, which had devised a new architecture called Explicitly Parallel Instruction Computing (EPIC) that combined the best of the other two architectures. The result was the Itanium.

So far, though, customers are not swarming to adopt the Itanium, which was formally launched last May after numerous delays and design problems. In the quarter ended Sept. 30, Gartner Dataquest estimates that Intel shipped about 2,162 Itanium chips, with 2,000 of those going to IBM for two big servers that use 1,000 Itaniums apiece.

``Itanium is not shipping where Intel wants it, not by any means,'' said Jef Hewitt, a Gartner Dataquest analyst. ``It won't be a world-shaker.''

Intel has always had low expectations for the first Itanium chip, but it has higher hopes for the next chip in the family, code-named McKinley, which is now being released in pilot systems.

Meanwhile, longtime foe AMD, a developer of Intel-compatible chips, has decided to extend the x86 CISC architecture into the 64-bit realm, going after the same market targeted by Itanium.

``All of the good ideas that made RISC run fast, people figured out how to put into CISC computers,'' said Fred Weber, AMD's chief technical officer. ``Time is clearly showing that changing the instruction set is not giving you any advantage in performance.''

While AMD is not expected to make inroads into the high-end server market, its Hammer chips, code-named SledgeHammer and ClawHammer, could further encroach on the desktop PC market, where AMD's Athlon has made big inroads in the past two years, and on the low-end server market, where Intel's Xeon chips have done well.

Intel's decision to back the novel Itanium architecture had upset a small group of Intel engineers in Oregon, who preferred to build on the x86 legacy. When AMD released the specifications of its upcoming 64-bit chips in the summer of 2000, these ``cowboy'' engineers decided that Intel needed to match its rival. They began developing their own 64-bit extensions to the Pentium line, making sure the code was compatible with AMD's design.

The Oregon team's initiative inflamed the feud between them and the Santa Clara team developing the Itanium, according to former engineers. But top executives, including Chief Executive Craig Barrett, reluctantly decided to support Yamhill development in addition to Itanium.

Intel executives knew the value of a backup plan from the company's own history: two decades ago, Intel developed a chip called the i432 that was supposed to reshape its future but failed miserably. Intel rushed out another chip, the 8086, which took just three weeks to design, and that chip became its bread-and-butter.

Whether the Yamhill features are ultimately activated depends on how AMD's new chips are received in the market. Kumar believes that Yamhill-enabled Pentium chips would help Intel fight off that threat from AMD without hurting sales of Itaniums for high-end servers. ``There can be parallel efforts,'' he said.

Yamhill sounds pretty good! Looks like Intel will remain the dominant chipmaker for server platforms.