Chip wars: Now it's best of three

Compaq's transfer of its Alpha chip technology to Intel means a shake-up in the competition for the high-end server niche. Eyeing the redrawn battlefield are the three remaining processor rivals: Intel, Sun and IBM.

And then there were three

Compaq Computer's announcement Monday that it will use Intel's Itanium chips in its high-end servers and transfer its Alpha chip expertise to Intel represents a dramatic overhaul of the market. As a result, just three major chip designs are left to compete: Intel's Itanium, Sun Microsystems' UltraSparc and IBM's Power.


Because Compaq's Alpha-based server business is in a distant fourth place compared with the Unix server lines from No. 1 Sun, No. 2 Hewlett-Packard and No. 3 IBM, Compaq's decision to rework its high-end server products is probably more important for Intel and Sun than for Compaq itself.

Among the effects:

  • Intel's Itanium efforts could benefit from new, experienced employees--but Sun and IBM also have an opportunity to snare Alpha designers.
  • The underpinnings of Alpha, UltraSparc and Power--a technology called reduced instruction set computing (RISC)--has just received a vote of no confidence, boosting the prospects of Itanium's very long instruction word (VLIW) design and putting more pressure on Sun.
  • Compaq's server-design employees will be able to unify products behind a single chip.

    "This makes an interesting dynamic, particularly for IBM and Sun being left as the only major RISC suppliers in the market," Gartner analyst George Weiss said Monday. "It raises more issues about whether RISC is in it for the long haul."

    Although IBM has hedged its bets by firmly backing both Power- and Itanium-based servers, Sun has derided Itanium as a "weird architecture" and insists its own UltraSparc design still has a glorious future before it.

    "All these architectures are converging gradually," said Shahin Khan, head of marketing for Sun's server products. In addition, Khan said, Sun will be able to improve its chips without requiring software companies to completely rewrite their software.

    On the other hand, Sun's competitors will be able to focus solely on designing servers, while Sun also must worry about designing the chips.

    "You wonder how long Sun will be able to do that before business or economics force their hand," Insight 64 analyst Nathan Brookwood said. RISC, which dates back to the 1980s, may have longevity problems. But Itanium, which was launched in May, hasn't even begun struggling through adolescence. The first generation of Itanium was delayed by years, and even Intel said the second-generation "McKinley" version will be the first with potential for use beyond just test servers.

    The deal with Compaq has the potential to improve that track record, though, because Compaq not only has expertise in chip design, but also in the creation of "compiler" software, which plays an increasingly important role in creating programs that take advantage of chip features.

    Under the deal announced Monday, Intel will incorporate Compaq technology to give an "essence of Alpha" to future Itanium designs, said Paul Otellini, general manager of the Intel Architecture Group.

    But some believe Intel won't get all it hopes out of the Compaq design staff.

    "Frankly, I don't think they're going to get much," said Linley Group analyst Linley Gwennap. "Most of the good Alpha engineers have already left. If they haven't already, most aren't going to be interested in working with Intel." The only other time Intel acquired a design team originally, from Digital Equipment--the 25-person StrongARM group under Rich Whitek--"the whole team walked out," Gwennap said.

    Regardless of design issues, Itanium likely will prevail on its business merits, Gwennap said, because so many server makers back the chip design. "At this point there's so much momentum behind the Itanium platform that it's hard to see how anybody's going to stand up against that," he said.

    One feature that could be useful for Itanium is "symmetric multi-threading," a process Compaq said would debut in the Alpha EV8 design. Symmetric multi-threading increases the efficiency of a chip by letting it work on more tasks in parallel. Another is better features on the chip for creating multiprocessor servers, said MicroDesign Resources analyst Cary Snyder.

    But symmetric multi-threading, like the VLIW design in Itanium, isn't much good without support from compiler software, Snyder said.

    Indeed, Intel is eager to put the Alpha compiler programmers to work. "We're hoping to get the compiler team (together) as quickly as possible," Otellini said. "They'll work on the overall (Itanium) compiler technology" and on bringing Compaq's operating Tru64, OpenVMS and NonStop Himalaya operating systems to Itanium.

    Sun also is interested in attracting some Alpha designers.

    "This strengthens what has already been a recruitment opportunity," said Khan, adding that Sun has chip design centers near Boston and Austin, Texas, where Compaq Alpha and IBM Power chips are designed, respectively.

    "We can bet that (Sun Chief Operating Officer Ed) Zander (and IBM COO) Sam Palmisano are going to vie for these engineers," Giga Information Group analyst Brad Day said. Compaq's experienced chip designers will be "naming their price."

    Cultural issues could be an obstacle preventing Intel from picking up all the Alpha designers, Snyder added. "What kind of toll the religious war will take remains to be seen. Now is a good time to motivate people with some really good low-cost stock options," he said. The change in Compaq's server designs is significant for its customers and for companies that sell software for Compaq servers. Switching from Alpha to Itanium will mean major work in programming and testing to see whether software and hardware work properly.

    The transition from MIPS chips to Alpha chips in Compaq's highest-end Tandem servers already has been a five-year process, with the first Tandem Alpha systems expected in 2003.

    But Compaq asserts the transition is worth it.

    When the company approached a number of large customers to test their reception to switching from Alpha to Itanium, the response "exceeded our expectations," said Mike Winkler, Compaq's executive vice president of global business units.

    "For the first time, they can see a constant migration and growth path that had always been a question--whether Alpha could continue to grow," he said. "What we found mattered to customers was not Alpha but the operating environment--Tru64 and OpenVMS. They didn't really care which chip."

    Sun, though, prides itself on how older software works on newer systems and asserts that Compaq's latest transition will create more sales of Sun servers.

    "This will accelerate (sales) of Alpha replacement systems," Khan said, predicting that future hurdles would afflict the transition to Itanium.

    Analysts--many of whom have been asking for years when the Alpha line will be phased out--were nearly universal in the opinion that it's a smart deal for Compaq.

    "In our view, Alpha was the albatross around Compaq's neck. And in this environment, it is just too much of a luxury to support," Buckingham Research analyst Peter Labe said in a research note Monday. "Compaq should save hundreds of millions in costs, focus its research more intensely, and simplify its go-to-market message to supporting just a single server architecture."

    Under the deal, Compaq will boost the speed of its current EV6 Alpha design twice more while preparing the release of its EV7 design. The EV8 team will work on Itanium instead, the companies said.

    "This is a commitment to ensure our customers...have a very clear, compelling road map that ends up in a great place," Compaq Chief Executive Michael Capellas said.

    The Alpha EV7 refresh "will be the end of the line," Winkler said. "We will then migrate to the Itanium architecture for all the Compaq server platforms."

    Weiss said the effect on Sun "wouldn't be evident until the 2003-2004 time frame. Just when depends on how quickly these technologies that Compaq contributes to Intel makes its way into the Itanium architecture. We don't believe that probably will happen until 'Madison,'" the chip due after McKinley. "So we're talking about 2004 or 2005 when Sun and possibly IBM need to be concerned about the growing dominance of Intel in the high-end server space."

    HP conceived of Itanium more than a decade ago and approached Intel in the early 1990s to build it so the design wouldn't be relegated to the small, high-end market its PA-RISC chips occupy. The company has been selling Unix servers since 1999 that can accommodate second-generation Itanium chips, and the last PA-RISC design, the 8900, is scheduled to be released in 2004, said Mark Hudson, worldwide marketing manager for HP servers. After that, it's Itanium all the way.

    HP is glad to have Compaq engineers to hasten the Itanium improvements and to impart more momentum to the chip. "Welcome to the club. This endorses what we've been saying for a long time: Itanium is inevitable," Hudson said. "It's the next-generation architecture for the next 15 to 20 years."

    The company is years ahead of Compaq in designing Itanium servers. But with Compaq's Intel server business much stronger than HP's at present, Compaq's change puts more pressure on HP's server division at a time when it is weak.

    "HP obviously worked closely with Intel on Itanium, and you would think they would be the preferentially treated vendor," said Technology Business Research analyst Lindy Lesperance. "But this turns everything around in favor of Compaq, which offers invaluable technology."

    Lesperance said that "Compaq's aggressive migration" will put "pressure on HP to expedite their Itanium migration strategy."