Intel's Huge Bet Turns IffyBy JOHN MARKOFF and STEVE LOHR
Itanium, a joint project of Intel and Hewlett-Packard, Silicon Valley's two largest companies, has been in the laboratory for more than a decade. Itanium is designed to excel at a sweeping array of advanced computing tasks, from solving grand scientific challenges to rendering complex graphics to slicing through vast databases. With more than 200 million transistors on each chip, it is designed to process data in big bites — 64 bits, in chipspeak — at blazing speeds. But Google isn't buying. And that is an ominous sign for what is one of the longest-running and most expensive computing projects in history — exceeded only by I.B.M.'s bet-the-company gamble on its 360 mainframes, which eventually succeeded. For Intel, Itanium's failure would be a painful black eye, a setback in its heavily financed assault on the corporate computing world beyond the personal computer. For its partner, Hewlett-Packard, the chip venture, if anything, is even more important. Exploiting the Itanium opportunity was crucial to the thinking behind Hewlett-Packard's $19 billion merger this year with Compaq Computer. If Itanium fails, it will be a severe blow. And if Itanium flops, Silicon Valley's psyche may well be shaken, too. Projects like Itanium are the valley's equivalent of going to the moon. Big budgets, big egos and a near-religious commitment to the "right way" to design a computer are involved. With Silicon Valley mired in its worst recession, some technologists there say that Itanium may be the industry's last such huge bet on computer design. Intel still says its commitment to Itanium is unwavering. Its smaller rival, Advanced Micro Devices, has an alternative to Itanium that computer makers are seriously considering. And Intel has a fallback project, called Yamhill, in case Itanium founders. But mainly, Intel is redoubling its bet on Itanium. It has taken an entire decade, an estimated $5 billion and teams of hundreds of engineers from the two companies to bring the first Itanium chip to market. As the struggles and costs mount for the companies, skeptical technologists say Itanium now has the hallmarks of a bloated project in deep trouble. It is already four years behind schedule, emerging just as companies are in no mood to spend money on technology. "Every big computing disaster has come from taking too many ideas and putting them in one place, and the Itanium is exactly that," said Gordon Bell, a veteran computer designer and a Microsoft researcher. INCREASINGLY, Intel is facing the risk that it has chosen the wrong path to high-performance computing. It may have looked backward as it developed the microchip equivalent of the behemoth computers of the past. Eric Schmidt, the computer scientist who is chief executive of Google, told a gathering of chip designers at Stanford last month that the computer world might now be headed in a new direction. In his vision of the future, small and inexpensive processors will act as Lego-style building blocks for a new class of vast data centers, which will increasingly displace the old-style mainframe and server computing of the 1980's and 90's. It turns out, Dr. Schmidt told the audience, that what matters most to the computer designers at Google is not speed but power — low power, because data centers can consume as much electricity as a city. If power efficiency does indeed trump processing speed, everything that Intel and Hewlett-Packard have done to pack raw power into the 221 million transistors of the new Itanium 2 could now be a handicap. The chip, which is as large as a silver dollar and whose 130 watts of power dissipation are enough to fry the proverbial egg, is not even a contender in the Google universe. "We're incredibly, incredibly power sensitive, and we've been talking to Intel about that," Dr. Schmidt said. So far, Intel and Hewlett-Packard have been hard-pressed to prove they are on the right track. The Itanium was announced in 1994 as a joint effort to design a processor for the world of large computing systems, like servers, mainframes and supercomputers. The project, however, has been beset by repeated delays. Originally planned to arrive in late 1997, the chip did not arrive commercially until last year. Worse, the first version of Itanium, code-named Merced, proved to be an embarrassing dud — its performance trailing even Intel's own 32-bit Pentium chips, let alone the rival chips it is intended to beat, like I.B.M.'s Power series or Sun Microsystems' Ultrasparc 64-bit processors. Itanium is Intel's effort to apply the economics of the personal computer business, with its lower costs of mass production and swift technological improvement, to the lucrative market for computer data centers of corporations and government. I.B.M. and Sun Microsystems, with their big server computers and 64-bit chips, are the leaders in the data center market — and the target of Intel and its partners from the personal computer industry, like Hewlett-Packard, Dell Computer and Microsoft. The more powerful Itanium 2 was introduced in July, but market researchers now project that Itanium will garner less than 10 percent of the market for server computing as far ahead as 2007. Intel executives have not disputed those forecasts. Yet Michael J. Fister, the veteran Intel manager who was called in to rescue the Itanium project three years ago, says the project has finally turned around. A little patience may be needed, he says, but what lies ahead for Itanium is opportunity, not insurmountable problems. Raised in Cincinnati and educated as an electrical engineer at the University of Cincinnati, Mr. Fister has worked at Intel for 15 years and was one of the managers behind Intel's successful 32-bit Pentium server line in the 1990's. He was chosen two years ago by Intel's chief, Craig R. Barrett, and its president, Paul S. Otellini, to reset the Itanium project. "Mike was the obvious choice," Mr. Otellini said. "He is one of our most seasoned design managers and an expert on high-end computer architecture." Intel has long been known as a magnet for sharp, hard-driving engineers, but Mr. Fister was a standout when he arrived in the 1980's, according to several people who worked at the company then. "I would always rely on him when I had a problem," said Jim Lally, a former Intel executive and now a partner at the venture capital firm Kleiner, Perkins, Caufield & Byers. "He would get it done and get it done right." Mr. Fister says Itanium is now on track. When Itanium 2 becomes available widely next year, he said, it will have better performance than any of the competing 64-bit chips. That, he added, will make up for all of Itanium's earlier failings. This month, at the Intel Developer Forum in San Jose, Calif., Mr. Fister detailed a road map for two new generations of Itaniums in 2003 and 2004. He spoke about a lower-power version of Itanium that might come closer to meeting Google's computing challenge. He even bragged about "10 tons" of Itanium-based computers the company was demonstrating at the conference. He acknowledged that Intel had spent too much time worrying whether its new 64-bit design would cannibalize its thriving Pentium processor market. Now that debate is over, he said. "We're seeding an architecture that is going to last for two decades," he added, even as he said critics were correct to be skeptical in the past. "We've been kind of plodding along and we missed a couple of schedule points," Mr. Fister said. But those days are behind Itanium, he said, adding, "The information technology industry is buying anything that demonstrates price performance." IN private conversations with big Intel customers, Mr. Otellini has said the company is more committed to Itanium than ever and willing to push to make it succeed — at whatever cost. A senior executive at a large computer maker said that when he asked about Intel's Itanium strategy, Mr. Otellini replied, "Don't even think about us pulling back from Itanium." The senior executive said his company still had concerns about Itanium — whether its complex design would work in the broad and diverse market of corporate computing. Mr. Fister argues that the debate about the complexity of writing software for the new chip design — a crucial issue if Itanium is to succeed — will soon be put to rest, as important software applications arrive for Itanium from companies like Microsoft, Oracle and SAP. Even if Itanium overcomes the software bottleneck, Mr. Fister will face another challenge soon: positioning Itanium against a new 64-bit Opteron from Advanced Micro Devices, Intel's closest rival in the personal computing world. Intel and Hewlett-Packard chose to start from scratch with a new design for Itanium, requiring programmers to rewrite applications. AMD has taken a more compatible approach. Opteron, which will begin shipping next year, is based on the original Intel-designed X86 instruction set. That means the chip will run all existing software intended for other Intel chips, as well as compatible processors, with only minor alterations. At a technical conference in Silicon Valley in the summer, computer designers were impressed when AMD said the new 64-bit extensions to the existing 32-bit chip required only 2 to 3 percent more silicon. That, they said, is a small price to pay for a 64-bit chip that can also run software applications written for existing 32-bit chips. "We believe customers want an easy migration path to 64 bits," said Marty Seyer, a vice president at AMD. Until recently, AMD has been largely ignored by Intel because it has not been a significant factor in the corporate computer server market. But if AMD finds success with a high-visibility computer maker like Dell, that could change fast. In fact, Dell has said that it is looking closely at using Opteron. "This is the first time AMD has had a value proposition that is more than just price," said Randall D. Groves, vice president for computer servers at Dell. "That's something we have to take seriously." Dell, he added, will probably make a decision before year-end on whether to buy AMD chips. The AMD challenge has led Intel to quietly begin a project, called Yamhill, in Oregon. There, a team of Intel engineers is trying to build a 64-bit extended version of its own X86 Pentium chips, just like AMD's Opteron. AMD's reputation in chip design is better than its reputation for manufacturing and business execution. But regardless of Opteron's fate, the Yamhill project itself could present thorny problems for Intel. If Yamhill shows an easier path from 32- to 64-bit computing than the costly rewriting of software applications, as the shift to Itanium requires, corporate customers will be reluctant to move to Itanium. Mr. Fister was coy about Yamhill in an interview at Intel headquarters in Santa Clara, Calif. Intel has never publicly acknowledged Yamhill's existence. But he suggested that AMD might encounter difficulties with Opteron. "It may not be as simple as people think it is to take advantage of a 64-bit processor," he said. Then, all but acknowledging Yamhill, Mr. Fister dropped a few intriguing hints that Intel's design efforts for it were well under way. "We're pretty smart about the evolution to 64-bit, and we're street-smart about it," he said. Mr. Fister also suggested that Intel might have already developed extensions to its 32-bit products that "we haven't told anyone about." But, he added, "it's hard for us to talk about unannounced entries." Not far away, in Palo Alto, Hewlett-Packard may have even more riding on Itanium than does Intel. The chip is a big part of the company's goal to gain an edge over I.B.M., Sun Microsystems and Dell in corporate computing. As a co-developer of the Itanium technology, Hewlett-Packard will get deep discounts on the price of each chip. That will give Hewlett a significant cost advantage over rivals. In an interview last December, before shareholders approved Hewlett's purchase of Compaq, Richard A. Hackborn, a board member and a former senior executive, portrayed Itanium as a potentially crucial advantage for the combined company. As part of its partnership with Intel, Hewlett was granted early access to chip designs and a deeper understanding of the technology, Mr. Hackborn said. So it will have a technical edge on other makers of Itanium-based servers, he argued, in addition to lower costs. That could make Hewlett the lead integrator of the Intel microprocessor and Microsoft operating systems in corporate data centers. "That's a huge opportunity," Mr. Hackborn said. If Hewlett can exploit that opportunity, he added, "That puts the new H.P. on a comparable footing with Intel and Microsoft." There are other benefits for Hewlett-Packard. The Itanium allows the company to eliminate both of its current 64-bit chips — the H.P. PA-RISC and Compaq Alpha. That alone should save the company $200 million to $400 million annually in development and manufacturing costs, according to Steven M. Milunovich, an analyst at Merrill Lynch. Yet if Itanium fails, he said, "the real loss is the opportunity cost." In the commercial market, Hewlett badly needs a credible strategy and a marketing message. Itanium, Mr. Milunovich said, is an essential part of the plan establishing Hewlett as the leading packager and integrator of Itanium-based systems. "If everything works according to plan, H.P. could be best positioned to be the company to take PC economics into the enterprise," Mr. Milunovich said. If Itanium fails, he added, Hewlett-Packard will be forced to go with Yamhill or AMD, and it will lose its hoped-for advantages in making large data-serving computers. In the end, Itanium may be most vulnerable to a force that neither Intel nor Hewlett-Packard can control: the economy. Even if Itanium proves a technical success, the most powerful incentive for companies to shift to it lies in the high-performance computing power needed for ambitious new information technology projects. But in the sluggish economy, few companies are increasing capital spending. "The real challenge to the Itanium may have less to do with marketing and design and more to do with a collapsing economy," said Michael Shulman, an analyst at ChangeWave, a research firm in Potomac, Md. Mr. Shulman's firm regularly asks corporate information technology managers about spending plans. From last summer to this summer, he said, interest in the Itanium eroded sharply as spending on new computing projects was reined in. "The common view is, `If it doesn't save us money we have no interest in it,' " Mr. Shulman said. "It's a very hard-nosed view."
|
OOGLE — the Internet's leading search engine, powered by an arsenal of computers with 15,000 microprocessors — should be a premier customer for 
They have done this before, but it wasn't the i860. That was the RISC chip, wasn't it? The earlier -- and very similar -- "breakthrough technology" debacle was the iAPX432, which when announced was going to be The Future Of Computing.