"I was asked if I would head a task force for one week on a very secret project and report directly to the vice chairman about IBM's possible entry into the telephone business, to be a flat-out rival of AT&T," said Joel Birnbaum, then head of experimental systems at IBM's research facility in Yorktown Heights, N.Y., and later the head of HP Labs, a post he held from 1984 until 1999.

The task force was to review the technology aspects of a proposed business partnership with a large European company aimed at delivering a family of digital telephone exchanges that would span small-office to giant carrier systems, leapfrogging AT&T with IBM's digital design clout.

Even after the European partner dropped out, IBM spent the next three and a half years building those systems. "The idea was that [the operation] would become as large as the existing IBM of about $30 billion," Birnbaum said.

But when the work was complete, IBM's top brass decided it would not take the risk of entering the new field. A group of 50 developers held an off-site meeting to review their options. They decided the highly scalable and simple telephone exchange processor developed as part of the project, later called the 801, had the potential to replace the processor elements used in IBM's flagship System/360 mainframes and other core systems. The group poured five more years' work into what is now generally seen as the development of the first RISC microprocessor, before IBM management again gave a thumbs-down.

"So when I came to HP [in 1980] I had it in mind to take the philosophy of RISC and build a distributed computer," Birnbaum said. Frustrated by the IBM experience, he had turned down an offer to head Big Blue's 600-person research group and joined HP, then a $3 billion computer and instrument company with a relatively fledgling 55-person computer science lab.

That was the genesis of HP's PA-RISC, a processor architecture launched in 1986. Combined with the company's internal HP-UX variant of Unix, it became one of the engines of HP's growth and the company's bet in the RISC/Unix wars of the 1980s and '90s. (HP has about $20 billion in computer revenue today, $20 billion in printer revenue and about $7 billion in services.)

RISC vs. X86

While PA-RISC family servers were on the rise, so was the Intel X86, commanding huge and ever-growing volumes that spread from desktops up to small servers, eating away at other RISC/Unix competitors such as MIPS and PowerPC. HP's management polarized around the RISC vs. X86 debate as separate RISC and X86 computer groups found themselves at odds.

John Young, HP's chief executive officer from 1978 to 1992, made an early decision to be No. 1 in PCs, a move that went largely unquestioned in the early days when desktops were profitable. But poor execution denied HP that top slot and the profits for which middle management fought for years.

Opening the door to the PC brought other changes, said a veteran HP engineering manager from the PA-RISC division, who has left the company and asked not to be named. Designing products and making a contribution are core values at HP, he said, but "the PC business brought us a new mentality. It's hard to innovate on a standard platform, so you don't try to make a contribution."

The tug-of-war between two competing computer cultures lasted until the mid-1990s, when Wim Roelandts and Rich Sevick, key executives shepherding the RISC/Unix business, departed HP for Xilinx Inc. The PC camp, led by Rick Belluzzo and Dick Watts, took over.

While Belluzzo is credited with helping grow HP's printer business, Unix group managers still hold a deep bitterness over his hard push for PC architectures — especially Microsoft's Windows NT, which was supposed to take Windows dramatically into new workstation and server markets. NT took several years to gain a strong footing in PC servers and never made anticipated inroads in workstations.

"He was way too much in love with Microsoft. He wanted to be their best partner," said the engineering manager. "So we sent confusing messages to customers while Sun came in and said, 'We are the Unix company,' and they got huge market share."

"It was a misstep," said a former HP microprocessor executive. "At that time people thought NT was going to take over the world, and the mainframe was supposed to die too."

(Belluzzo later left HP to become chief executive of an ailing Silicon Graphics Inc. There he quickly pushed the company to transition away from its proprietary MIPS microprocessor and standardize on NT and Intel's upcoming Itanium. But SGI got burned when Itanium was delayed by two years, a lag Belluzzo might have anticipated from reports from microprocessor engineers co-developing the architecture who worked for him at HP. In February, Belluzzo was named president and chief operating officer of Microsoft Corp.)

"Now the HP Pavilion [consumer desktop PC] is the No. 1 in retail, but there's no profits left. There was no pot of gold when we got to the end of the rainbow," said the engineering manager.

Very long design cycle

In 1993, researchers from HP Labs in Palo Alto presented the systems design group in Cupertino, Calif., with a next-generation processor architecture that became the essence of the company's current technology bet and the seed of a unification between its warring PA-RISC and X86 camps. The Wide Word architecture was the brainchild of Josh Fisher and Bob Rau, who had pioneered ideas in very long instruction word (VLIW) processors in separate startups before joining HP.

But the Cupertino team insisted HP lacked the market clout to establish such a radical new architecture. HP decided that rather than fight Intel as MIPS and PowerPC partners had done, it would try to engage the big chip maker. So near Thanksgiving of 1993 Lew Platt, who had become HP's chief executive in 1992, called Intel chief executive officer Andrew Grove to discuss a partnership that would make the new 64-bit Wide Word architecture the follow-on both to the PA-RISC and the Intel Pentium.

"It was a very short meeting," is how Birnbaum recalled Grove's rejection. Roelandts later approached Grove again with the same proposal, but with more details about providing the key backward compatibility Intel wanted with the X86.

Grove was hooked enough to agree to a "bake-off" between Wide Word and a competing architecture Intel had been developing internally. By early 1994, an intensive year-long negotiation began on the partnership that was finally announced, in early 1995, as the IA-64.

"We didn't have to invest in a fab. We were guaranteed compatibility with PA-RISC. And we were guaranteed that all the most important software companies would be writing apps onto our architecture," said Birnbaum, describing the advantages of owning a piece of a microprocessor with Intel.

Perhaps most important, HP gets special discounts on the price of Itanium chips from Intel. While the exact discount and its duration have never been revealed, it's likely to be one of the driving forces in the HP-Compaq merger. If the company gets the lowest price on IA-64 chips and also sells the greatest volumes of systems based on them, its financials would outstrip just about any other server maker including Sun, which designs all of its processors, and IBM, which designs and fabs some of its server processors.

But so far execution has hampered theory. The initial IA-64 chip, dubbed Itanium, faced more than two years of delays due in part to its ambitious mandate of creating a mainstream VLIW-like architecture that is backward compatible with X86 applications and systems software. A tapeout slated for 1997 came in July 1999. Systems that were to ship in 1998 came out in late 2000 — and met underwhelming response due to the microarchitecture's lackluster performance.

The companies had split the design work. Intel took the lead on the first-generation Itanium, working with a handful of HP engineers. HP devoted about 100 engineers in its Fort Collins, Colo., operation to the next generation, dubbed McKinley, now due out in 2002. "In retrospect it might have been better to do one project, not two, and get to market faster," said the former IA-64 executive.

Once again, while HP was distracted — this time on IA-64 — Sun was quick to jump into the Internet fray, reaping heady gains in the 1999-2000 boom as HP chugged along at a more modest pace.

"It's been a big disappointment here that it's taken so long for IA-64 to come out," said Birnbaum, who is now a technical adviser to HP CEO Carly Fiorina. "The idea was McKinley would be out in 2000 at latest. We've done all kinds of shrinks and tricks to keep PA-RISC alive. We had hoped not to have to do too much of that."

The more potent implementation of IA-64, expected in the McKinley chip, might give the partners their first taste of success in what has stretched into a 10-year effort.

"That's when you will tell if this was successful or not," said the former IA-64 executive, who has worked closely with both companies. "It's a big-stakes, big-risk, big-rewards game and the outcome will affect the lives of tens of thousands of people who build chips, systems and software. But it will take a couple more years until the verdict comes back."

Birnbaum says the story amounts to a 30-year effort to realize a vision of complex servers as part of a back-end information utility, while PCs and their follow-ons create an environment of pervasive computing. Indeed, Birnbaum claims he coined the term "information appliance" back in 1975.

"All of this stuff started long ago," said Birnbaum, who added that he is not involved in the current merger talks with Compaq Computer Corp. "HP has been on a track for a long time, but we lost our way. Now the final chapter is about to be played out. It could be this merger is exactly what is needed to bring this vision to life. Or it could be [that] even though it's a good idea, it will die in execution."