(FORTUNE Magazine) – The collaboration between Intel and Hewlett-Packard to develop the Merced, a turbocharged big brother to the ubiquitous Pentium microprocessor, has been anything but stealthy. Everybody in Silicon Valley has known about the project since Intel and HP announced it matter-of-factly back in 1994.

Still, salient details about the Merced, named after a river that flows out of Yosemite Valley, have been scant. Outsiders knew only that the chip, like several other high-end microprocessors already on the market, would crunch data in 64-bit chunks, vs. 32 bits for the Pentium, and that it would ship in commercial quantities sometime late in the decade. Ironically, the very scarcity of information about the Merced created more buzz than dweeby Intel and HP probably could have stirred up even if they had hired Steve Jobs to promote it.

Hence the air of rapt anticipation in mid-October at the Microprocessor Forum, an annual Silicon Valley gabfest for serious chipheads, where Merced's principal designers finally revealed just how ambitious--and how revolutionary, if you fully believe the hype--Merced will be. This usually dry event was suddenly the hottest ticket in town. More than 1,200 attendees, including dozens of journalists, were packed into the ballroom at San Jose's Fairmont Hotel as densely as transistors on a DRAM. Scores of others were turned away.

Merced's masterminds--John Crawford of Intel and Jerry Huck of HP--addressed a lofty presentation to their peers, spending lots of time on the arcana of state-of-the-art microprocessor design and dwelling on topics like "memory latency" and "predication and speculation." They didn't reveal very much that a normal PC user would relate to--like how much the chips will cost, how many transistors they'll contain, or how fast they'll run. But there was no mistaking the import of what Crawford and Huck described. When one of Intel's $5 billion semiconductor fabs starts pumping out Merced chips in 1999, they will embody a significant shift in microprocessor design. Just as important, if the Merced measures up to expectations, it could all but eliminate what serious competition Intel still faces in the lucrative general-purpose microprocessor business.

Intel is already sitting pretty, holding an 80%-plus share of the market for microprocessors that go into PCs. That dominance has made Intel one of the most profitable companies on earth: For the nine months ended in September, it made $5 billion on sales of $18 billion, and appears to be on track to rack up more than $25 billion in annual sales. But margins have shrunk slightly in recent quarters because of competition from Pentium clones; a new high-end processor would give Intel a boost by broadening its markets.

HP, for its part, is a big shot in the computer business, one of the top five PC makers and racing neck and neck with Sun Microsystems to be No. 1 in Unix computers. For the nine months ended in July, HP reported profits of $2 billion on sales of $31 billion. Still, the computer market is ferociously competitive. With Merced, HP hopes to stand out as a technology pioneer in what is becoming a commodity business.

Crawford and Huck picked a suitably grandiose acronym to describe their design approach: EPIC, short for Explicitly Parallel Instruction Computing. The design will let a single Merced act like a whole squadron of processors handling perhaps more than a dozen computational operations simultaneously, vs. four parallel operations in today's hottest chips. The chip will also be able to run today's Windows software, as well as programs written for HP's version of Unix, all without any modification.

While Intel and HP wouldn't comment, experts think the Merced could perk along at three times the "clock speed" of top-of-the-line Pentiums, or roughly 900 million cycles a second. The combination of speed and parallel processing translates into performance that by 2000 should let beefier cousins of today's PCs take on even the most demanding computing tasks. Says Intel CEO Andy Grove: "The whole point of Merced is to bring the microprocessor revolution to the big iron of enterprise computing. It is the one remaining segment that hasn't felt the full benefits of microcomputing." In other words, watch out, mainframes.

In the past 18 months, Merced has taken on greater significance and urgency than Intel anticipated. That's because the emergence of the Internet has ignited the market for network servers, basically Unix workstations or hefty PCs outfitted with scads of memory and phalanxes of disk drives. Many big corporations have opted for Unix servers made by Sun, Digital Equipment, HP, or IBM, largely because they incorporate microprocessors much more powerful than Intel's top-of-the-line Pentiums, the engines of PC servers. What's more, rather than wait for PC servers to pack more horsepower, customers are beginning to shift heavy corporate computing tasks to these Unix machines. Intel felt the need to take the wraps off Merced less to impress the world with its technological prowess than to convince potential customers that the chips will be worth the wait.

Marketing concerns aside, to hear Crawford and Huck tell it, EPIC is the third great milestone in microprocessor history since Intel engineers invented the gizmos back in 1971. The first microprocessors, of which the Pentium is a direct descendant, employed Complex Instruction Set Computing, or CISC, architecture. CISC chips hardwire as many as 200 logical and mathematical operations in silicon, a repertoire that can handle the most exotic computation. (Think of a business calculator with 200 functions and a button for each.)

The second wave came in the mid-1980s, when makers of engineering workstations switched to so-called Reduced Instruction Set Computing, or RISC. As the name implies, RISC chips work using fewer operations, dropping complex functions that programmers rarely tapped. (Think of a calculator with a few dozen buttons that can be combined to perform a wide range of functions.) This relative simplicity makes RISC chips easier to design and smaller and cheaper to build. They are also inherently faster than CISC chips, and in the chip business, speed means power. The best-known RISC chips are Sun's Sparc, HP's PA (Precision Architecture) RISC, Silicon Graphics' MIPS, and Motorola's and IBM's PowerPC.

The Merced will employ elements of CISC and RISC, plus extra circuits that enable it to run lots of parallel operations. So much for the technical advance. What really sets Merced apart from other 64-bit chips is something Crawford and Huck didn't boast as much about: Intel's manufacturing prowess. Because Intel pushes the envelope in its fabs, it can make chips with far smaller transistors than most of its rivals. This lets the engineers cram more transistors on a chip and arrange them in novel ways.

Crawford and Huck and the Merced team have used this freedom to lay the groundwork for a chip with mind-boggling market potential. They've endowed Merced with the ability to run Microsoft and Unix programs--software in which customers have already invested billions of dollars and upon which their businesses depend. To maintain compatibility with today's Windows programs, a sector of the chip will be devoted to the stripped-down core of a Pentium. Likewise, to guarantee that existing HP-Unix programs will run, another tiny quadrant will hold PA-RISC circuits.

Other areas of Merced will house the new EPIC circuitry, including a gang of processing units, each a 64-bit-RISC cell no bigger than the [F] logo at the end of this story. These will let Merced run many operations in parallel, once software is written to tap them. (Intel and HP won't say how many cells the first Merceds will incorporate.) Merced's processors will all share on-board memory and control circuits.

Merced's true power lies in its triple personality, argues a competitor who should know. Vinod Dham, former chief of Intel's Pentium group and now leader of the microprocessor group at Advanced Micro Devices, the leading maker of Pentium clones, says, "The real achievement and beauty of Merced is not so much its ability to run many operations in parallel, but the fact that it can do all that and still be 100% compatible with today's Windows and HP-Unix software. I don't mean this disparagingly, but a more accurate acronym than EPIC might have been MOTS, for More of the Same."

That's not to say Merced's internal parallelism isn't a remarkable achievement. Interestingly, the most crucial part of the EPIC juggling act is accomplished not on the chip itself, but long before a program even is loaded into it. When a software developer writes a program, it must be "compiled" into "machine language," a code that a microprocessor can understand and act on. (The packaged software you buy has already been compiled.) In the months leading up to the release of the first Merced chips in 1999, Intel will supply programmers with special compiler software that will not only translate their work into code for Merced but also parse the programs for sequences of instructions that don't depend on results from other sequences. These "independent instructions" can run in any order and be carried out by whatever part of the Merced chip is available. In a typical program, more than half the instructions are independent.

Moreover, the Merced has been designed so that hundreds of the chips can be yoked to handle gargantuan computing tasks that have historically been the province of mainframes and supercomputers--tasks like managing huge databases of credit card transactions or modeling weather systems. With Merced, Intel will have, for the first time, power to compete across the full spectrum of computing.

Sounds impressive, and it is. Still, the question remains: If Intel is so rich and has so much design expertise and such clear manufacturing advantages, why did it need a partner to develop Merced? And why HP?

To answer the first question, it helps to remember that back in 1994, when Intel decided it needed help, IBM--the No. 1 computer company--and Motorola--the No. 2 microprocessor maker--had joined forces and were about to deliver a hot new RISC processor called the PowerPC. Apple was the initial customer for that chip, but IBM hinted that the PowerPC was central to its own computer plans as well. PowerPC was able to get to market in a hurry largely because several companies shared in designing the chip for specific computers, developing compilers, and testing software that would tap its power. In short, PowerPC had Big Mo.

Intel had its own 64-bit chips on the drawing board, but worried that unless it got help from a company used to building and selling computer systems, it might miss the mark. Intel had little experience designing general-purpose RISC chips, and RISC features would be needed in any chip that hoped to run lots of operations in parallel.

HP seemed a natural candidate. For one thing, it and Intel are Silicon Valley neighbors. Both are known for corporate cultures that celebrate engineering. HP had also recently decided to get serious as a PC maker and was becoming a big Intel customer. Most important, HP was producing its own highly regarded RISC processor--the PA family--and had some of the world's leading authorities on microprocessor design working in its labs.

HP officials, meanwhile, were doing serious soul-searching. As proud as they were of their PA chips, they knew the cost of producing new generations would soon become prohibitive (1994 was the year that the price tag for a topnotch chip fab passed the $1 billion mark). Moreover, since HP was in the business of making both Unix servers and conventional PCs, its officials worried about the expense and complexity of supporting two product lines based on fundamentally different hardware designs. So when Intel made an alluring offer, HP leaped at the chance: Why not work together to build a new generation of microprocessors that could run both Unix and Windows, and that in the future would take advantage of leading-edge designs hatched in HP's labs?

As it turned out, PowerPC wasn't the dire threat Intel feared. But despite some early friction, the more Intel and HP worked together, the more the partnership made sense. Recalls HP's Huck: "Intel is a little more confrontational than we are, and that took getting used to. But all in all, we complemented each other very well."

The benefits to HP are obvious. At a time when the computer industry is moving away from proprietary systems toward hardware and software standards, HP has intimate ties with the leading microprocessor maker. The relationship nicely complements HP's other major alliance, a deal with Microsoft announced earlier this year, in which HP will, among other things, provide service and support for some of Microsoft's biggest enterprise customers. Moreover, Merced should rejuvenate HP's $6-billion-a-year Unix computer business at a time when HP might otherwise be tempted to narrow its focus to computers running Windows.

Intel benefits on several levels too. HP helped it come up with a chip that really does advance the art of microprocessor design, one that should be able to outperform all comers in terms of raw performance. Moreover, Intel will finally get a solid foothold in Unix servers and workstations, a pesky industry segment with fiercely loyal customers that for years has consistently managed to deliver superior microprocessors and computing power. By marketing Merced as a high-end complement to the venerable Pentium--Crawford says Intel has plans for at least six more versions of Pentium that should carry it well into the next decade--Intel will have a complete line of chips ready to power anything from the lowliest laptop to a supercomputer. Eventually Merced will begin to displace Pentiums in mainstream desktop and laptop PCs. Grove doesn't expect wholesale replacement until 2004, but Merced gives Intel the luxury of thinking that far ahead--and beyond. AMD's Dham is only half-joking when he says: "Merced will ensure Intel's dominance of the microprocessor business for the next 50 years."

What will Merced do to competing chips? Some rival products seem to be weakening, and Merced isn't even in production. The PowerPC is flagging along with Apple. The same goes for the MIPS and Silicon Graphics. Intel now is deciding whether to assume stewardship of Digital's Alpha microprocessor as part of a settlement of a legal dispute. Even Sun, the fiercely independent Sparc-chip maker currently locked in a nasty feud with the other half of the Wintel duopoly (namely Microsoft), plans to provide a version of Solaris, its Unix operating-system software, to support the Merced. CEO Scott McNealy won't rule out the possibility of even adopting the chip at some point. He notes that NCR has already announced that it will build Merced-powered systems that use Solaris.

This being the computer industry, there could always be a hitch, of course. The last time Intel broke new ground with a mainstream microprocessor was in 1985, when it introduced the 80386, a Pentium precursor that was the company's first 32-bit processor. The whole point of the 386 was to give IBM and Microsoft a platform for an advanced 32-bit operating system that tapped this new power. When did Microsoft deliver software that took full advantage of 32-bit chips? Fully ten years later, when it introduced Windows 95.

Grove and Crawford say that won't happen this time: Microsoft has been working on a 64-bit version of Windows NT for Merced for a couple of years and has become much more proficient. Don't hold your breath, though. As everybody knows, software is hard. If Microsoft really does manage to deliver a specially tailored version of Windows NT at the same time that Merced computers come out, well, that will be a truly revolutionary event.

REPORTER ASSOCIATE Alicia Hills Moore