The startup that saved ATI

The startup that saved ATI
Rick Merritt
4/21/2003 11:06 AM EDT

No one expects that ATI Technologies Inc. (Markham, Ontario) will
drive graphics king Nvidia Corp. completely off its pedestal. But the
scrappy No. 2 player in 3-D chips for PCs is staging a comeback of
sorts, at a time when computer graphics are making an architectural
shift to programmability and a market shift to integrated parts.



Over the past couple of years, ATI has overhauled itself with
management and engineering prowess from acquisition ArtX. Now, most
observers expect it will re-emerge as Nvidia's equal, helping set the
pace in both PCs and game consoles as a maturing graphics industry
heads into what many say will be uncharted waters.


ATI is poised to launch at Computex next month what could be the
first integrated chip set to run Microsoft's DirectX 8.1 API, use dual-
channel 400-MHz double-data-rate memory and link to the 800-MHz bus
for Intel's next-generation Pentium, dubbed Prescott. The chips could
leapfrog anything chip set makers Intel, Silicon Integrated Systems
and Via can offer in graphics while also beating Nvidia, which makes
chip sets for AMD CPUs but has no Intel processor bus license.


The launch is especially sweet for ATI, which lost its perch atop PC
graphics after Intel Corp. rolled out a new category of chip sets with
integrated graphics in late 1999. Afterward, the erstwhile market
leader saw its share of desktop sockets nearly halved.


But before it faded into graphics history, ATI beat out offers from
competitors S3 Corp. and Nvidia to acquire ArtX (Palo Alto, Calif.),
seen by many as the last hot startup in the maturing PC graphics
industry. ATI bought ArtX in February 2000 for about $400 million in
stock and options.


Since that time, the 70-person startup has been reinvigorating ATI
with efforts that are now starting to bear fruit on several fronts.
"The center of gravity for ATI has definitely shifted from Canada to
California," said one observer who asked not to be named.


The story begins in late 1997, when a handful of top engineers and
managers from Silicon Graphics Inc., many of whom had helped design
the Nintendo 64 console, got an idea for a startup. They would cram
high-end graphics into a PC chip set and leapfrog the giants of the
mainstream desktop world by leveraging what they learned from
designing a high-performance, low-cost game box.


The result was ArtX, which got initial funding from Taiwanese PC
maker Acer Inc. About nine months later, old contacts from Japan came
seeking a partner for their next-generation console, which later
became the Nintendo GameCube.


"They had given up on SGI. The last of the people they trusted were
gone, and they went looking for the people. It's not a company-to-
company thing for them; it's a person-to-person thing," said Greg
Buchner, at that time the head of ArtX.


The visit sparked a debate at the small startup. "We said we really
didn't want to divert ourselves, but Nintendo can make a pretty
compelling argument and it was a pretty huge opportunity, so we
decided to go ahead in mid-'98," said Tim Van Hook, chief designer for
the Nintendo 64 and a founder of ArtX. So ArtX forged a deal to
develop the Flipper chip for the console code-named Dolphin in return
for royalties. "We knew we couldn't take on the [chip] manufacturing.
That would require as many more people as we had in the whole company
at that time," said Joe Macri, another SGI veteran who became the 23rd
person to join ArtX. He is now a director of technology at ATI.


At Comdex/Fall in 1999, the startup launched with some fanfare its
ArtX1 PC chip set. By that time, the company had hired as its
president David Orton, a hard-charging former manager of SGI's
advanced-graphics division, who was keen to take ArtX public. However,
an IPO looked risky. As it turned out, the Comdex splash brought the
company lots of attention-and acquisition offers from ATI, Nvidia and
S3.


It wasn't hard sorting out those bids. S3 was already in trouble and
would break up in April 2000. "We could see the initial signs of
that," said Buchner. As for Nvidia, "we didn't think it would work
culturally or from a valuation perspective."


ATI was the clear fit. It was trying to get its own integrated-
graphics program off the ground to catch up with Intel, which was
wreaking havoc with the market. ATI had an Intel bus license, but it
had no presence in the console space, no office in Silicon Valley and
was badly in need of a makeover. Indeed, ArtX and ATI managers
separately described ATI at the time of the acquisition as "a sea" or
"a blob" of engineers without clear lines of responsibility. "They
were a startup with one big organization," said Buchner.


In what turned out to be a case of the tail wagging the dog, ArtX's
Orton was named president and chief operating officer of the merged
entity from the outset. He reorganized ATI into separate business
units and three major design teams under a handpicked set of managers
who shared his drive to compete.


"He is someone who loves a good fight and he loves to win it," said
Buchner, now one of two chief technology officers and four vice
presidents of engineering at ATI.


Leveraging the ArtX team in Palo Alto, Orton created a Silicon Valley
base for ATI just a mile down the road from Nvidia's sprawling green-
marble headquarters in Santa Clara. Engineers at the ATI site finished
the GameCube graphics chip, then led the design for the R300 graphics
core, ATI's first to execute Microsoft's DirectX 9 application
programming interface.


The DirectX spec was driving a new architectural direction in PC
graphics. Rather than delivering fixed functions based on
approximations using integer math and a graphics pipeline pioneered by
SGI, DirectX 8.1 had taken a new course: toward more general-purpose
programmable vector processors based on more-exacting floating-point
calculations.


Ultimately, it is thought that the DirectX evolution will lead chip
makers to create devices based on dozens of computing elements that
can calculate polygon vertices and run pixel-shading programs for a
variety of graphics and video applications. Sony, IBM and Toshiba
apparently share this vision. Their Cell architecture-announced in
March 2001, though not yet released-could someday use hundreds of
cores in a parallel array to power future Playstation consoles and a
wide variety of other broadband multimedia products.


"It's all about programmability now. That's the new battleground,"
said Peter Glaskowsky, editor of the Microprocessor Report. "These
chips are not distinguished by the number of parallel pipelines or
clock rates anymore. The key issue is how much can you do to each
pixel you draw, how many programmable instructions you can run per
pixel."


In this environment, ATI tacked into the wind. Unlike past cores that
aimed for good-enough graphics, the R300 represented an effort to
match or beat the best desktop chip Nvidia might offer. "We didn't say
get the best performance at 10 x 10-mm die size. We just said get the
best performance. We had to go out and capture the flag," said Orton.


And last August, ATI did just that, launching its Radeon 9700 six
months before Nvidia shipped its GeForce FX part for DirectX 9. That
lead-a rarity in PC graphics, where new cores ship every 12 to 16
months-was as much a triumph of execution for the reorganized ATI as
the result of stumbles at Nvidia.


According to Macri, the tale of the tape fell on two strategic
decisions. Nvidia opted for a 128-bit memory bus linked to next-
generation GDDR-2 memory and made in the latest 130-nanometer copper
process. ATI chose a 256-bit-wide memory bus geared for fast transfers
over existing GDDR memory and made in a 150-nm technology-effectively
the last generation of aluminum interconnects.


"You can blame me for the 150-nm decision," said Buchner, who leads
on silicon issues in his CTO role. "It was one of the biggest unending
arguments in the company. It was not a question of 130 nm not being
ready. It was more about trying to hit the ground running and asking
how many risks we want to take."


For its part, Nvidia attributed its delays on the GeForce FX to
"getting the chips to yield at speed," said Dan Vivoli, vice president
of marketing. Nvidia recently signed IBM as a fab partner in addition
to its existing one, Taiwan Semiconductor Manufacturing Co. Some think
that move could give it an edge getting to 90-nm chips.


While the high-end graphics chips get much of the attention, chip
sets with integrated graphics have swept across the market, becoming
in some ways more strategic. Market watcher Mercury Research
(Scottsdale, Ariz.) estimates that by the start of 2003, as much as
half of all desktops used integrated graphics, a category that didn't
exist before 1999. In this sector, ATI's progress has come more
slowly.


ATI combined its internal teams with those from ArtX and another
acquisition, Chromatic Research. After misfiring on a couple of
generations, ATI aimed at its core notebook users and hit the jackpot
with the U1, which shipped last summer. ATI's integrated chips now
sell at about a million units per quarter, a notch above Nvidia's
integrated parts, said Dean McCarron, principal at Mercury Research.
The company will pitch the integrated chip set that is to debut in May
for low-cost, high-performance consumer systems. That positioning
might be a sop to soften competition with technology partner Intel,
which could continue to command the chip set space for business
desktops that don't require heavy graphics.



Also next month, Microsoft Corp. will make a move that will likewise
strengthen ATI's hand in chip sets. At the Windows Hardware
Engineering Conference, Microsoft will detail plans to use a 3-D
interface on its next version of Windows, dubbed Longhorn and slated
for 2005. Such a move could bolster ATI as one of the few chip makers
capable of easily rolling out a DX 9 chip set in time for Longhorn's
release. A higher profile for 3-D could ultimately boost the fortunes
of all the graphics companies and set the stage for the PC, not the
console, to command the most attention among software developers.


"When the OS has 3-D as part of the environment, that's the point
when 3-D moves into everybody's world," said Dave Rolston, who heads
ATI's 175-person Silicon Valley office.


In this expanding environment, the ATI vs. Nvidia battle has lots of
legs. Nvidia will launch within days its next-generation core, the
NV-35, which is expected to sport a 256-bit memory bus and other major
enhancements. "It will be unambiguously the best," said Vivoli.


Meanwhile, both companies have hit the market with versions of their
latest DX 9 parts aimed for mainstream desktops, where most of the
money in PC graphics lies. "The design wins are happening right now,"
said analyst McCarron.


Beyond that the two are set to slug it out all over again with their
next-generation cores built for the new PCI Express interconnect. ATI
is also challenging Nvidia for the graphics socket in the next-
generation Microsoft Xbox and is charging into consumer applications
such as cell phones and set-top boxes. "In 2004 ATI will become a
visual-computing company beyond the PC. We've got to get into a faster-
growing part of the market," Orton said.


http://eetimes.com/electronics-news/4044411/The-startup-that-saved-ATI

 

I didn't realize April 1 had arrived early!

 

ATI? Weren't they known for a product called the VGA Blunder?