August 27, 1997
Intel and Hewlett-Packard
Gamble With Faster Chips
By JOHN MARKOFF
AN FRANCISCO -- Some of the nation's top computer chip designers
are scratching their heads about what Intel and Hewlett-Packard are
up to in a top-secret project to design a new chip -- code-named
Merced.
|
Illustration: Christine M. Thompson / CyberTimes
|
The Merced, set for release within the next two years,
represents both an opportunity and a risk for Intel. It would mark
the most significant shift in the history of the immensely popular
Intel microprocessor family, which began in 1979 with the Intel
8088 chip that powered the first IBM PC and has continued through
Intel's latest, the Pentium II.
The Merced would be a departure from that tradition because the
chip would use an "instruction set" radically different from the
one that has evolved in Intel microprocessors.
The instruction set, etched into the chip's circuitry, comprises
hundreds of basic operations, like adding, subtracting, multiplying
and moving, which all other components and software in the computer
depend upon.
Changing the instruction set for personal computing is akin to
introducing a new industry alphabet -- with all the potential for
grammatical lapses and communications breakdowns such a fundamental
shift suggests.
That change, on top of the perhaps seven-fold increase in the
number of transistors packed onto the surface of the chip, portends
a greater transformation in the computer industry in the next five
years than in the preceding 25 years.
The first Intel personal computer chip in 1979 had 30,000
transistors; the most recent Pentium II has more than 7 million.
When it arrives, the Merced is expected to contain 20 million to 50
million transistors, said John Novitsky, a former member of Intel's
Pentium's chip design team who is now vice president of marketing
for the PC component maker Micro Module Systems.
And the chip will probably have a basic clock speed of almost
1,000 megahertz -- more than twice the raw performance of today's
fastest chips. Moreover, the Merced will be a 64-bit microprocessor
-- compared with the 32-bit limit of the current Intel Pentium. That
means the computer can process twice as much information at once,
promising much faster and complex searches of data bases and more
realistic audio and video multimedia capabilities.
The consensus in Silicon Valley is that if Intel, the world's
largest chip maker, fumbles the shift to the new chip family, it
risks damaging its near stranglehold over the personal computer
hardware business.
No wonder the engineers who build the various other chips and
hardware components in today's PCs -- all of which depend on being
compatible with Intel's technology -- are intensely curious about
what new features Intel may be planning. (Intel and Hewlett-Packard
have had little to say on the subject so far. While Hewlett is a
partner in the development, most computer executives agree that
Intel is driving the project.)
"It's a good bet that Merced will mean a rather drastic
change," said John Wharton, a former Intel microprocessor designer
who is now an independent chip designer and who moderated a panel
at Stanford University on Monday attended by some 500 of Silicon
Valley's leading engineers. In the absence of guidance from Intel
itself, the group pondered the industry-quaking question, "What
would I do if I were designing the Merced?"
It is not a question for geeks alone. Investors, too, are always
alert to any vicissitudes involving Intel, as was evident last
Friday when Intel's shares plunged $6 dollars before recovering to
end the day down $2.1875, closing at $96.1875.
The dip came after Thomas Kurlak, a well-known semiconductor
analyst at Merrill Lynch, cut his rating on Intel to "neutral"
from "buy," and reduced his 1998 profit estimate for the chip
maker to $4.90 from $5.60.
Kurlak is forecasting an industry slowdown in 1998, as
corporations cut their capital spending, and he predicts Intel will
face increasing pressure on its profit margins.
The stock price has continued slipping in the last two trading
days, ending Tuesday at $92.6875, down $1.50.
Already, there have been signs of increasing pressure on Intel,
which is based in Santa Clara, Calif., as Advanced Micro Devices,
Cyrix and other chip makers who make clones of the Intel product
line have continued to harry the company with increasingly
competitive products. But in a broader sense, Intel has become
increasingly dominant in the microprocessor business in recent
years.
For example, both Sun Microsystems and Digital Equipment offer
highly sophisticated alternatives to Intel's microprocessor
technology. But Sun and Digital have tiny market shares in chips
for PCs, work stations and so-called network servers compared with
the more than 64 million microprocessors for such machines that
Intel sells each year.
And the one PC chip that represents a true consumer-market
alternative to the Intel standard -- the Power PC used in the Apple
Macintosh and Macintosh clone machines -- has an uncertain future.
Work on the next-generation Power PC by a team of developers from
Apple Computer, IBM and Motorola has stalled, industry engineers
say.
So, in a very real sense, the future is Intel's to control or to
fumble. Which is why the chip designers at the Stanford panel were
so intent on pondering the imponderable.
For one thing, the Merced's designers must figure out how to
insure that the chip's new instruction set will be able to read and
run the thousands of existing MS-DOS and Windows software programs
written for Intel's older family of chips -- or risk having more
than 80 percent of the world's existing PCs be rendered suddenly
obsolete.
And yet, the designers will not want to hamstring the Merced's
potential to perform at speeds not possible on earlier Intel chips.
Most computer designers now expect the new Merced to be
something of a silicon chameleon -- able to adapt to and handle many
different types of software written for other computer-chip
instruction sets, whether for older Intel chips or competing
hardware.
But that adaptability, computer designers say, could be the
greatest danger. If because of its flexibility the new Merced runs
existing Microsoft software slower than can some current Pentium
chips, the market may be slow to embrace it. The fact is that
little new software is likely to be initially available that can
take advantage of the Merced's next-generation capabilities.
In the early going, in other words, the Merced may simply seem
musclebound to most users. Hence the risk of Intel's planned
revolution.
"They're sitting on the most successful computer architecture
in history," said Keith Diefendorff, an Apple Computer chip
designer who appeared on the Stanford panel. "I'd milk the current
Pentium architecture for another 25 years. There is no reason to
change anything."
Related Sites
Following are links to the external Web sites mentioned in this article. These sites
are not part of The New York Times on the Web, and The Times has no
control over their content or availability. When you have finished visiting any
of these sites, you will be able to return to this page by clicking on your
Web browser's "Back" button or icon until this page reappears.
