TAKING ON INTEL
How companies like AMD, National
Semiconductor, and IDT are trying to
snatch a slice of the microprocessor
market from Intel.
By Michael Slater
The Red Herring magazine
November 1998
It's hard to imagine that less than 20 years ago Intel was
an innovative but small maker of semiconductors. IBM's
selection of Intel's 8088 microprocessor for its new
personal computer in 1981 forever changed Intel--not to
mention the entire landscape of the semiconductor and
computer industries. Adeptly seizing the opportunity
created by the explosion of the PC market, Intel became
the chip industry's powerhouse; it earned $6 billion on
$24 billion in revenues last year. Only three companies in
the world--Exxon, General Electric, and Royal
Dutch--enjoyed greater profits in 1997.
The 100-million-unit market for PC processors has lured
half a dozen Intel competitors, led by Advanced Micro
Devices (AMD), onto the field, but profits have been
elusive for these companies. With nearly 85 percent
market share in units and a much higher share of the total
revenues, Intel doesn't allow its rivals much room in
which to maneuver. It owns the majority of the world's
high-performance microprocessor fabs, or manufacturing
plants. It is the driving force behind the evolution of the
PC's architecture, the dominant supplier of system-logic
chip sets, and an emerging force in graphics processors.
With nearly $8 billion in cash, it can afford anything from
glitzy marketing campaigns to intensive technology
development. Keeping up with Intel on Intel's terms is a
battle that no company can win.
But a market this big has niches that no one player, not
even Intel, can fill. The alternatives to Intel's processors
are attractive for one reason: they cost less. Intel's
competitors--the publicly held AMD, National
Semiconductor's Cyrix division, and Integrated Device
Technology (NYSE: AMD, NSM; Nasdaq: IDTI) or
privately held startups like Rise Technology, Transmeta,
and Metaflow (now owned by
STMicroelectronics)--typically make lower-end chips
that they price at 25 percent to 50 percent below Intel's
comparable products; they can't compete at all with
Intel's most expensive processors. Of course, at the low
end the profit margins are also less.
The rapid rate of price declines and performance
increases in PC microprocessors makes staying in
business even harder for Intel's competitors. Price
declines of 25 percent per quarter are routine, and Intel
cut the price of some of its chips more than 40 percent
between June and September. To keep their processors'
average selling prices from plunging, companies must
keep introducing higher-speed versions--and customers
thus expect substantially more for their money every few
months.
Aside from price, the key factor in creating opportunities
for Intel's competitors is the desire of some of the largest
PC makers, including Compaq, IBM, Hewlett-Packard,
Acer, and Packard Bell, to break free of the chip giant's
grip. These companies would like to procure
microprocessors the way they procure other PC
components: through a competitive process in which they
are free to choose the most cost-effective parts.
Moreover, they don't want their brand subordinated to
Intel's brand.
The surge of interest in sub-$1,000 PCs during the past
18 months has been a godsend for Intel's competitors.
Because buyers who focus on paying the lowest price for
a PC are less brand-sensitive than buyers of higher-end
systems, PC makers are more willing to put a non-Intel
processor into an entry-level machine. Lower price
points have been most important in the retail channel,
allowing non-Intel processors to become a big part of
this segment. In the United States five of the top six retail
PC vendors offer systems with AMD or Cyrix
processors; no longer are non-Intel chips associated
primarily with off-brand systems. According to ZD
Market Intelligence, AMD or Cyrix processors were
used in 60 percent of the sub-$1,000 desktop machines
and in 38 percent of all x86 desktop PCs sold through
the U.S. retail channel in June.
On the business side, however, buyers are traditionally
more conservative, less focused on price, and more
conscious of brand. PC makers thus have been wary of
using alternative processors in their lines of business PCs.
But with the dramatic success of these chips in consumer
machines, penetration of the business-PC market is likely
to follow. And follow it must, if Intel's competitors are to
grow as much as they hope to.
AMD: FROM INTEL FAMILIAR TO CHIEF RIVAL
Many companies carry the imprint of their leader's
personality, but few do so more than Sunnyvale-based
AMD, Intel's chief rival. Led by Jerry Sanders, a
flamboyant salesman with a passion for beating Intel,
AMD was founded in 1969, a year after Intel was
started. Like Intel, AMD is a traditional semiconductor
company with its own fabs. From the start, though,
AMD's focus was on building veritable clones of chips
already offered by other companies but with some
feature benefit, like lower power usage or higher speed.
After years of literally duplicating Intel's designs through
patent cross-licenses and other agreements, AMD set off
on its own to produce a Pentium-class processor. Its first
independent creation was the K5, which lagged in speed
and was late to market. In a bold attempt at a quick
recovery in late 1996, AMD acquired NexGen, a
struggling startup whose 586 processor had modest sales
and which was coming dangerously close to running out
of cash. NexGen's executives then took over the
leadership of AMD's PC microprocessor efforts.
The NexGen 686, already in prototype form at the time
of the acquisition, became the AMD K6. AMD was
widely criticized for having paid too much for NexGen,
but considering the immense short-term importance of
the K6 chip and the NexGen team to AMD, the
company may have made an excellent deal. But although
the K6 has rescued AMD's PC processor business from
oblivion, it has not generated profits so far. The company
shipped 2.7 million K6 chips in the second quarter of this
year, representing more than 10 percent of the overall
x86 PC processor market in units but at an average price
of only $86, compared with more than $200 for Intel's
equivalent. AMD's ultimate goal is to grab 30 percent of
the market.
The key to AMD's near-term future is the K6's
successors: the K6-2, which began shipping in June, and
the K6-3, scheduled to ship by the end of the year. The
K6-2's most significant feature is a new set of
instructions, called 3DNow, that speeds 3D graphics and
is AMD's biggest feature advantage over Intel to date.
AMD hopes that the K6-2 will drive a marked increase
in its average selling price in the second half of the year.
In 1999, however, as faster versions of Intel's new
Celeron processor, with larger on-chip caches, set the
pace for entry-level PCs, the K6-2--even with its
enhanced graphics capability--will be relegated to use in
bargain-basement systems.
Most of the Intel-alternative chips, including AMD's K6
family, use souped-up versions of Socket 7, the same
interface (which connects the processor to the rest of the
system) that Intel used for its Pentium. Intel, meanwhile,
has all but abandoned the Socket 7 design; for the
Pentium II, it switched to an interface called Slot 1. So
far no competitor offers a processor compatible with this
interface. The Slot 1 design has some performance
advantages, the most important of which is that it has a
separate bus for cache memory, which gives the
processor much faster access to the cache than is
possible in today's Socket 7 chips.
AMD's K6-3 will add more cache memory to the
processor chip, offering the same benefit as the
dedicated cache bus in Intel's Slot 1 design. If AMD
delivers the K6-3 as promised, the question then will be
whether the company can secure enough customers to
increase its shipments significantly. This will be a
challenge because Intel is seeking to shift the entire
market away from the Socket 7 interface and will offer
increasingly compelling Celeron processors for
entry-level systems.
AMD's strategy beyond the K6 has been seriously
complicated by a concession the company made in its
negotiations with Intel: in return for a renewal of its patent
cross-license, AMD agreed that Socket 7 was the last
Intel processor interface that AMD would duplicate. This
has forced AMD to find an alternative to the Slot 1
interface for its Pentium II-class chips. The company's
solution is something it calls Slot A, which borrows its
electrical design from Digital Equipment's latest Alpha
chip. The Slot A processor module will require a
motherboard and chip set that is different from that of
any Intel processor.
For the development of its upcoming K7 processor,
then, AMD faces both a significant challenge and
substantial opportunity in parting from Intel's design.
AMD must establish an entire chip set and motherboard
infrastructure--and this time, unlike with all its previous
processors, AMD's chip will not be compatible with PC
system designs created for Intel's processors. If AMD
can deliver all the pieces, however, the Slot A strategy
could succeed; PC buyers have no reason to care about
Slot 1 or Slot A as long as their systems work well. With
the K7, AMD may finally be able to compete across the
breadth of Intel's product line--if it can convince PC
makers to use the Slot A design and if it can overcome
the brand preferences of high-end PC buyers.
NATIONAL SEMICONDUCTOR: REDIRECTING
CYRIX
Until its acquisition by National Semiconductor last fall,
Cyrix was the scrappy kid on the block. A
venture-funded startup based in Richardson, Texas,
Cyrix broke into the business by starting with so-called
math, or floating-point, coprocessors. The company had
both hits and misses, but it has never been consistently
profitable.
Unlike Intel and AMD, Cyrix was until recently
dependent on other companies to manufacture its chips.
It has had a number of foundry partners, but most of its
chips have been built by IBM's microelectronics
division--and the relationship has been stormy. National
is now starting to build Cyrix chips in its own fab.
Cyrix's MII chip plugs into the Socket 7 interface and
performs comparably to a Pentium II on typical business
applications. It is slower at graphics processing,
however, than the Pentium II and K6-2. This
shortcoming has forced Cyrix to price the MII very
aggressively; it is by far the cheapest chip in the
300-MHz class.
Inexpensive as the MII has become, Cyrix has a different
solution for the lowest-end systems: the MediaGX, the
only PC processor today that combines CPU and
peripheral functions. By offering a CPU, memory
interface, PCI interface, and graphics controller all on
one chip, the MediaGX reduces overall system cost. The
chip's CPU and graphics speeds are slow, but Cyrix
plans to address the deficiencies in its MXi chip, which is
due out by early 1999.
National has steered Cyrix to concentrate on its highly
integrated chips for the low-cost PC and
consumer-device markets and to avoid direct
competition with Intel. To this end, National is
developing a PC-on-a-chip, promised for mid-1999,
that goes beyond the MediaGX to add more peripheral
functions. The chip will be too slow for even low-end
PCs but could be useful in information appliances like
Web access terminals, for which performance
expectations are lower. National plans to make a faster
but low-end PC-on-a-chip based on Cyrix's MXi in
2000. As National molds Cyrix to deliver low-cost,
highly integrated chips for information appliances, the
future of Cyrix's processors for mainstream PCs is in
doubt.
IDT: FROM SRAMS TO LOW-COST
PROCESSORS
The newest competitor to begin shipping x86 processors
is Integrated Device Technology (IDT), an established
$600 million developer of SRAMs, specialty memories,
and embedded processors that it builds in its own fabs.
Seeking a higher-margin product line, IDT funded a
startup design house, Centaur Technology, in 1995.
Centaur, which operates as a wholly owned subsidiary in
Austin, Texas, brought its first processor to market in
two years and introduced a second design less than a
year later.
IDT seemed to enter the x86 business with more humility
than other competitors. From the start it focused on
developing low-cost chips. Its designs are the industry's
most compact, minimizing manufacturing cost. But the
recent average selling price for the company's PC
processors has been below $45--lower even than IDT
had intended. Furthermore, it has shipped only a few
hundred thousand units per quarter, perhaps 1 percent of
the market. But by steadily improving its products, IDT
hopes to become a leading manufacturer of low-cost PC
processors.
Indeed, IDT is already well along in its goal, with three
new designs in development for its WinChip processor
line. They are intended chiefly to serve the Socket 7
market that Intel has abandoned and that Cyrix and
AMD will deëmphasize in 1999. By the middle of next
year, IDT may well offer the industry's fastest Socket 7
processors while competing with Cyrix to power the
least-expensive PCs.
IBM: LOOKING FOR PARTNERS
IBM's microelectronics division, although relatively low
profile, is nevertheless an important player in the PC
processor business. That IBM has made money on PC
processors where others have failed is due in large part
to its unique business model--it has paid nothing for the
chip designs. This exceptional arrangement came as part
of its agreement to serve as Cyrix's foundry. IBM would
commit chip-building capacity to Cyrix at favorable
prices; in return, IBM could sell, without paying royalties,
as many Cyrix-designed chips under its own name as it
shipped to Cyrix.
But with National Semiconductor, Cyrix no longer needs
IBM to build its processors. Cyrix's and IBM's strategic
interests also have diverged: IBM wants to focus on
higher-performance processors; Cyrix, on low cost.
Unfortunately for IBM, it decided several years ago to
redirect its internal x86 design team to focus on the
PowerPC--the once-promising alternative to x86
processors (see "A Requiem?" May 1997, page 78).
For x86 design, IBM chose to rely on partners like Cyrix
and NexGen. Now that Cyrix is part of National,
NexGen is part of AMD, and the PowerPC is limited to
the small Macintosh market, IBM needs a new source
for x86 designs. To meet this need, the company is
rumored to be involved with other x86 design houses,
including Rise Technology and Transmeta. IBM also will
begin making chips for IDT late this year, supplementing
that company's in-house fab capacity.
THE UPSTARTS: RISE, TRANSMETA, AND
METAFLOW
The lucrative PC processor market continues to draw
new contestants. Rise Technology of Santa Clara is
closest of these to shipping a product; its mP6 chip is
due out this fall. The 80-person company has disclosed
few details other than that it will initially target makers of
low-cost notebook computers. (At our press time Rise
was planning to unveil the design of its mP6 in October
at Microprocessor Forum, where Intel, AMD, Cyrix,
and IDT also will make new disclosures.)
Headed by chairman and CEO David Lin, previously a
manager in NEC's processor operation, Rise has raised
funding from venture capitalists, investment bankers, and
PC manufacturers that the company declines to name.
Nor will it disclose the foundry partner that will build its
chips, but IBM is a good bet.
Another secretive--yet strangely well known--Santa
Clara startup is Transmeta, headed by former Sun
research manager and processor architect Dave Ditzel.
Transmeta hasn't been able to conceal its existence
entirely but has refused to divulge anything about its
plans. The company has more than 100 employees,
including high-profile technologists like Colin Hunter, an
expert in translating x86 programs to other architectures;
Linus Torvalds, creator of the Linux operating system;
and Robert Collins, best known for his Intel Secrets
Web site (www.x86.org). Sources who insist on
anonymity indicate that Transmeta's chip has its own
native instruction set and uses software translation to
provide complete x86 compatibility. The company is not
expected to start production on the chips until 1999.
Transmeta appears to be more performance-focused
than the other Intel alternatives, but it is not yet clear
which market segment it will target. The company admits
to being venture funded, and it also took an investment
from Microsoft cofounder Paul Allen; its foundry
partners are rumored to be IBM and Toshiba.
Also a potential Intel competitor is Metaflow
Technologies of La Jolla, California, a pioneer in
advanced microprocessor designs. The company was
acquired last year by STMicroelectronics, an early
partner of Cyrix that has not produced PC processors
for the past few years. ST presumably intends to return
to the business with Metaflow's x86 chip design.
THE ODDS
Of all Intel's competitors, AMD is clearly placing the
biggest bet. With one large fab in Texas dedicated to PC
processors and a new fab in Germany due to open in
1999, AMD has invested billions in manufacturing
capacity. And to succeed with the upcoming Slot A
interface and K7 chip, AMD must develop more
infrastructure than ever before.
For AMD to achieve its desired 30 percent market share
would require most other Intel competitors to fail. Four
or five companies will fight AMD for the role of Intel
alternative in certain parts of the market. Cyrix, IDT, and
Rise are focused on entry-level PCs, where price is more
important than brand; AMD hopes to rise above this
battle and serve the midrange to high-end market.
Cyrix's strategy is as risky as AMD's, but in different
ways. By focusing on very low cost PCs and information
appliances, Cyrix hopes to avoid direct confrontation
with Intel. On the other hand, this focus leaves Cyrix
dependent on a market that is just emerging--and where
the profit per chip is small and high volume thus essential.
IDT's plan is the most straightforward: provide low-cost
processors that are compatible with Intel's prior
generation. IDT's modest investment in its x86 business
and its compact design will be advantages in the
competition with AMD--albeit producing a smaller
return.
Ultimately, Intel is a manufacturing company, and it has
made huge investments in its plants. The company is
strongly motivated to keep these fabs full; if necessary,
Intel is likely to compromise its profit margins to keep its
market share from slipping too far. This is a terrifying
prospect to Intel's competitors, whose primary edge is
lower prices. If the PC market grows more quickly than
Intel's manufacturing capacity, the company will focus its
efforts on developing more expensive processors, and
many suppliers could prosper. But if the market gets
weaker, or if all the growth is at the low end, Intel will
have to compete more aggressively on price--and the
results could be very ugly for everyone else.
Michael Slater is the founder of MicroDesign
Resources, executive editor of Microprocessor
Report, and program director for Microprocessor
Forum.
Table of Contents
|
