Personally, I think it was the ATOT effect.
FEW COMPANIES have closer ties to Intel Corp. than Hewlett-Packard Co. After
all, the two Silicon Valley giants collaborated for a decade on chip technology
that H-P plans to use in three major computer lines.
Yet H-P also has become a vocal fan of Opteron, a competing microprocessor
from Advanced Micro Devices Inc. H-P executives concluded in the summer of 2003
-- after developing a prototype computer under a secret effort called Project X
-- that AMD's chip was the best for some server systems, and Intel didn't have
a plausible strategy to close the gap.
"We kept asking Intel, `What are you going to do for us to be competitive?' "
says Paul Miller, H-P's vice president of marketing for industry-standard
servers. "The plan kept on falling short."
Many plans have fallen short lately for Intel, which struggled with a series
of execution problems in 2004. But the biggest image problem is AMD's perceived
performance edge in the chip technology Intel invented. Though AMD's stock
plunged yesterday because of Intel's counterattack in chips called flash
memory, AMD's transformation from a technology follower to a leader could
weaken Intel's long-held hegemony over computing standards.
How did AMD, an erratic performer through much of its 35-year history, make
such big waves? Part of the answer can be traced to Intel decisions in the
1990s that might have outflanked its smaller rival but wound up having the
opposite effect.
Intel bet billions of dollars on the chip technology proposed by H-P, a sharp
break from the x86 design that is the electronic brain of most personal
computers. But the chip Intel designed, called Itanium, arrived years late and
has become popular only in a narrow class of high-end servers. Intel's x86 Xeon
and Pentium chips, by contrast, became the dominant calculating engines in
computer workstations and high-volume servers.
Intel last month said it would hire H-P's chip-design team and take over all
Itanium development, ending the joint effort begun in 1994. H-P says it will
invest $3 billion over three years on computers and software associated with
the chip.
Itanium remains a hot topic in Silicon Valley, derided by competitors and
defended by its creators. It also set off fierce internal debates at Intel,
which former company engineers say contributed to delays in improving the x86
line. AMD tried to exploit the situation.
"I knew that there were people inside of Intel whose careers were tied to
Itanium," says Fred Weber, AMD's chief technology officer. "They had nothing to
gain by allowing something to compete with it internally."
The battle is linked to a technology transition that means little to most
users now but may play a big role in how computers evolve. The shift, to chips
that process 64 bits of data at a time, allows computers to use much more
memory than today's 32-bit machines.
Itanium was designed to run new 64-bit software; it also runs 32-bit programs,
but more slowly than expected. AMD's Opteron chip, by contrast, is a modified
x86 chip that is good at running both 32-bit and 64-bit programs.
AMD's approach, launched by its former chief executive, Jerry Sanders, and
pushed by successor Hector Ruiz, is crucial to its penetration of the corporate
market. AMD accounted for only 4.3% of x86 server sales in the third quarter,
the research firm IDC says. But H-P, Sun Microsystems Inc. and International
Business Machines Corp. are Opteron customers; even Dell Computer Inc., long an
Intel-only customer, has said it might try Opteron.
Intel has responded by overhauling its product plans, including adding
Opteron-like 64-bit technology to its Xeon chips -- an indignity for a company
used to defining hardware standards.
"Maybe we should have reacted a little faster," says Abhi Talwalkar, a vice
president in charge of Intel's enterprise-platforms group. But he adds, "I
think we've responded incredibly fast over the course of the last 12 months."
AMD, of Sunnyvale, Calif., long has made Intel-compatible chips. But Intel
insisted that AMD be barred from emulating Itanium under a 1996 agreement
between the companies, recalls Atiq Raza, a Silicon Valley investor who then
directed AMD's development efforts. He says Intel also declined to share
information about a technology for connecting microprocessors with other chips.
Forced to innovate, AMD engineers added 64-bit circuitry to an x86 design.
They also devised a simpler way to fetch data from memory, boosting Opteron's
32-bit performance. "It was almost as though Intel closed doors for us that
allowed us to find new ways to compete with them," Mr. Raza says.
Such technologies long had been discussed inside Intel. But some former Intel
engineers say top managers resisted changes to x86 chips that might suggest a
weakening commitment to Itanium, then called Merced.
Robert Colwell, an architect on three x86 chips before resigning in 2001, says
his boss let it be known he would be fired if Mr. Colwell kept lobbying to
enhance x86 chips to exploit more memory -- a hallmark of Merced. Albert Yu, a
former Intel executive who was then Mr. Colwell's boss, denies any threat and
says he was eager to improve x86 chips, too. "I certainly was encouraging him,
not trying to discourage him," he says.
In 1998, several high-ranking engineers met with Chief Executive Craig Barrett
to discuss their concerns about Merced, two participants say. After listening
politely, Mr. Barrett abruptly ended the meeting, accusing them of wanting to
kill a competing project, they say.
Mr. Barrett wasn't available for comment. Chuck Mulloy, an Intel spokesman,
calls the account of Mr. Barrett's behavior "unlikely." But he adds: "Intel has
very talented, creative engineers. There are disagreements all the time."
AMD's courtship of H-P began in 2001, in a server group that was part of
Compaq Computer Corp. until the 2002 merger of H-P and Compaq. Teams drove
regularly between AMD's Austin facilities and Compaq's offices in Houston, as
H-P analyzed AMD's technology, manufacturing capacity and marketing strategy,
H-P's Mr. Miller says.
In late 2002, the effort known as Project X was established to design an
Opteron-based server. Engineers were sequestered in a special office so they
wouldn't violate Intel restrictions about use of its technical data. "We put
these guys in a glass bubble," Mr. Miller says.
When the first machine was booted up in mid-2003, the performance ranged from
40% to four times better than Intel-based machines, Mr. Miller says.
Still, H-P didn't announce its relationship with AMD until February of 2004 --
after the announcement of Intel's 64-bit Xeon and H-P's commitment to also use
that chip. H-P plans to keep using such x86 servers for high-volume
applications and Itanium for high-end chores, says Rich Marcello, who runs
H-P's business critical-systems group.
Cliffnotes:
Intel messed up
AMD came out with Opteron
AMD now in the CPU performance lead
FEW COMPANIES have closer ties to Intel Corp. than Hewlett-Packard Co. After
all, the two Silicon Valley giants collaborated for a decade on chip technology
that H-P plans to use in three major computer lines.
Yet H-P also has become a vocal fan of Opteron, a competing microprocessor
from Advanced Micro Devices Inc. H-P executives concluded in the summer of 2003
-- after developing a prototype computer under a secret effort called Project X
-- that AMD's chip was the best for some server systems, and Intel didn't have
a plausible strategy to close the gap.
"We kept asking Intel, `What are you going to do for us to be competitive?' "
says Paul Miller, H-P's vice president of marketing for industry-standard
servers. "The plan kept on falling short."
Many plans have fallen short lately for Intel, which struggled with a series
of execution problems in 2004. But the biggest image problem is AMD's perceived
performance edge in the chip technology Intel invented. Though AMD's stock
plunged yesterday because of Intel's counterattack in chips called flash
memory, AMD's transformation from a technology follower to a leader could
weaken Intel's long-held hegemony over computing standards.
How did AMD, an erratic performer through much of its 35-year history, make
such big waves? Part of the answer can be traced to Intel decisions in the
1990s that might have outflanked its smaller rival but wound up having the
opposite effect.
Intel bet billions of dollars on the chip technology proposed by H-P, a sharp
break from the x86 design that is the electronic brain of most personal
computers. But the chip Intel designed, called Itanium, arrived years late and
has become popular only in a narrow class of high-end servers. Intel's x86 Xeon
and Pentium chips, by contrast, became the dominant calculating engines in
computer workstations and high-volume servers.
Intel last month said it would hire H-P's chip-design team and take over all
Itanium development, ending the joint effort begun in 1994. H-P says it will
invest $3 billion over three years on computers and software associated with
the chip.
Itanium remains a hot topic in Silicon Valley, derided by competitors and
defended by its creators. It also set off fierce internal debates at Intel,
which former company engineers say contributed to delays in improving the x86
line. AMD tried to exploit the situation.
"I knew that there were people inside of Intel whose careers were tied to
Itanium," says Fred Weber, AMD's chief technology officer. "They had nothing to
gain by allowing something to compete with it internally."
The battle is linked to a technology transition that means little to most
users now but may play a big role in how computers evolve. The shift, to chips
that process 64 bits of data at a time, allows computers to use much more
memory than today's 32-bit machines.
Itanium was designed to run new 64-bit software; it also runs 32-bit programs,
but more slowly than expected. AMD's Opteron chip, by contrast, is a modified
x86 chip that is good at running both 32-bit and 64-bit programs.
AMD's approach, launched by its former chief executive, Jerry Sanders, and
pushed by successor Hector Ruiz, is crucial to its penetration of the corporate
market. AMD accounted for only 4.3% of x86 server sales in the third quarter,
the research firm IDC says. But H-P, Sun Microsystems Inc. and International
Business Machines Corp. are Opteron customers; even Dell Computer Inc., long an
Intel-only customer, has said it might try Opteron.
Intel has responded by overhauling its product plans, including adding
Opteron-like 64-bit technology to its Xeon chips -- an indignity for a company
used to defining hardware standards.
"Maybe we should have reacted a little faster," says Abhi Talwalkar, a vice
president in charge of Intel's enterprise-platforms group. But he adds, "I
think we've responded incredibly fast over the course of the last 12 months."
AMD, of Sunnyvale, Calif., long has made Intel-compatible chips. But Intel
insisted that AMD be barred from emulating Itanium under a 1996 agreement
between the companies, recalls Atiq Raza, a Silicon Valley investor who then
directed AMD's development efforts. He says Intel also declined to share
information about a technology for connecting microprocessors with other chips.
Forced to innovate, AMD engineers added 64-bit circuitry to an x86 design.
They also devised a simpler way to fetch data from memory, boosting Opteron's
32-bit performance. "It was almost as though Intel closed doors for us that
allowed us to find new ways to compete with them," Mr. Raza says.
Such technologies long had been discussed inside Intel. But some former Intel
engineers say top managers resisted changes to x86 chips that might suggest a
weakening commitment to Itanium, then called Merced.
Robert Colwell, an architect on three x86 chips before resigning in 2001, says
his boss let it be known he would be fired if Mr. Colwell kept lobbying to
enhance x86 chips to exploit more memory -- a hallmark of Merced. Albert Yu, a
former Intel executive who was then Mr. Colwell's boss, denies any threat and
says he was eager to improve x86 chips, too. "I certainly was encouraging him,
not trying to discourage him," he says.
In 1998, several high-ranking engineers met with Chief Executive Craig Barrett
to discuss their concerns about Merced, two participants say. After listening
politely, Mr. Barrett abruptly ended the meeting, accusing them of wanting to
kill a competing project, they say.
Mr. Barrett wasn't available for comment. Chuck Mulloy, an Intel spokesman,
calls the account of Mr. Barrett's behavior "unlikely." But he adds: "Intel has
very talented, creative engineers. There are disagreements all the time."
AMD's courtship of H-P began in 2001, in a server group that was part of
Compaq Computer Corp. until the 2002 merger of H-P and Compaq. Teams drove
regularly between AMD's Austin facilities and Compaq's offices in Houston, as
H-P analyzed AMD's technology, manufacturing capacity and marketing strategy,
H-P's Mr. Miller says.
In late 2002, the effort known as Project X was established to design an
Opteron-based server. Engineers were sequestered in a special office so they
wouldn't violate Intel restrictions about use of its technical data. "We put
these guys in a glass bubble," Mr. Miller says.
When the first machine was booted up in mid-2003, the performance ranged from
40% to four times better than Intel-based machines, Mr. Miller says.
Still, H-P didn't announce its relationship with AMD until February of 2004 --
after the announcement of Intel's 64-bit Xeon and H-P's commitment to also use
that chip. H-P plans to keep using such x86 servers for high-volume
applications and Itanium for high-end chores, says Rich Marcello, who runs
H-P's business critical-systems group.
Cliffnotes:
Intel messed up
AMD came out with Opteron
AMD now in the CPU performance lead
