Project X: VM Lab's parallel-processing MPEG-2 core [Wired]

Rohit Khare (
Wed, 14 Oct 1998 21:02:29 -0700

Excerpted from :

=A0=A0F E A T U R E =A0| =A0Issue 6.07 - July 1998

=A0Inside Project X

The digital living room continues to elude companies like Sony and
Microsoft. VM Labs, a Valley unknown, has the answer: a platform for all
home entertainment. The game is total market penetration.

By Charles Platt


This explains the strange behavior of stock-option millionaires who continu=
working 10 hours a day, seven days a week, peering into screens "for the fu=
of it." Their gratification isn't dollar denominated; it derives simply fro=
disseminating their code among as many PCs as possible. That's the ultimate
validation in this insanely competitive intellectual endeavor: total market

Alas, in an industry dominated by large corporations, brilliant software
sometimes languishes in obscurity and lone developers have a tough time
stealing market share. In the words of Richard Miller, who founded an
obscure start-up named VM Labs in January 1995: "The real question was, How
could I possibly come up with a world-beating new interactive product
without being Microsoft or Sony?"

Twice before, Miller had tackled radical, ambitious business ventures - wit=
unhappy results. In his British homeland, he played a major role in
developing the world's first laptop, the Z88, at Cambridge Computer, founde=
by legendary PC pioneer Clive Sinclair. The Z88 contained just four chips:
one for RAM, one for ROM, another for the processor, and one that Miller
designed for all the other computational functions. It was a tour de force,
a masterpiece - and a commercial disaster.

Undaunted, Miller founded Perihelion Inc. with a modest ambition: Compete
with Sun Microsystems. "We developed the world's first parallel-processing
graphics workstation," he explains, "a networkable graphics processor, so
you could build a board with 50 of them running simultaneously." Atari
licensed it, and it too failed commercially.

For a while, Miller surrendered his autonomy and became an Atari employee.
But in November 1994 he quit, sensing a once-in-a-lifetime opportunity. A
video-compression standard named MPEG-2 had just been announced, creating
fantastic possibilities. Miller believed it could enable him to compete on
an equal basis with industry giants, catalyzing a new wave in home

MPEG-1 used this system, and in China it's still the compression scheme of
choice, driving the sale of almost 15 million VideoCD players in 1997 alone=
but almost all the content is pirated, because Hollywood never accepted the
format. In the United States, MPEG-1 was a failure.

At the end of 1994, most observers doubted that MPEG-2 would do better. It
was more powerful but required 2 Mbytes of RAM, which seemed excessive for
consumer electronics devices. True, MPEG-2 was being adopted by DirecTV, th=
satellite system - but in 1994 no one knew whether that would succeed,
either. DVDs were going to use MPEG-2, but even they were an iffy
proposition, since everyone knew that laserdiscs had never displaced dull
old videotape.

In fact, the reception for MPEG-2 was so lukewarm that no one bothered to
develop new silicon to decode its compressed format. Instead, DVD
manufacturers planned to use a kluge of existing general-purpose chips that
would be relatively inefficient and expensive. Richard Miller figured he
could design a single custom chip, just as he had for the Z88 laptop, that
would do the job faster. But the real payoff would come if he made his chip

Today's DVD players are dumb devices. All they do is read discs and display
pictures. Miller imagined enhancing a DVD machine with a multipurpose chip
like a CPU, enabling many additional functions. It should be so flexible, s=
programmable, he thought, that it would run game software and do 3-D
modeling with unprecedented realism.
Thus, a DVD player could evolve into something like the mythical sub-$500
home computer. It would serve as an all-new hardware platform for a new
generation of entertainment software, and since it would plug into any TV
set, it could reach millions - maybe hundreds of millions - of homes around
the world. Moreover, Miller hoped that by sneaking this new processing powe=
into the living room, he might catalyze a significant shift in consumer
tastes: the often-predicted, never-realized maturation of videogames into
adult-friendly interactive entertainment.

In March 1995, Miller obtained seed funding. In June he hired his first
employee. By the end of the year he had half a dozen people, all sworn to
secrecy. A while later, with a fine sense of drama, they started wearing
mysterious black windbreakers with "VM Labs" embroidered in silver on the
front pocket and "Project X" splashed in garish colors across the back.

The firm rented space at the rear of a two-story cream stucco building in
Los Altos, on San Antonio Road. For two years hardly anyone knew what VM
Labs was up to, and the people who did know were skeptical, because the pla=
was audacious to the point of lunacy. Miller wanted his chip in every DVD
player, every digital TV, every videogame system, every dedicated Web
browser, and maybe in some videophones and some PCs as well. He would reach
out and touch the TV-oriented mass audience, doing an end run around
companies such as Microsoft that had been trying and failing to achieve the
same goal.

Today, as the chip enters the market, Miller actually has a shot at
realizing this goal.

In a small conference room stands a vintage arcade videogame: Defender, by
Williams. Software boxes and volumes of documentation have been stacked
haphazardly on the kind of bookshelves you can get cheap at Office Depot. A
not-very-clean whiteboard occupies the far wall, and some utilitarian gray
formica folding tables stand in the center of the room.

At one of these tables sits Bill Rehbock, vice president of third-party
development, a jovial and rotund character with a youthful face, like an
overgrown schoolboy. He has serious glasses, a calculator wristwatch, and a
pen in the pocket of his gray short-sleeved shirt. He's the personification
of Thirtyish Computer Nerd. But he also projects an aura of calm, practiced
competence. When he speaks, he's definitive and categorical, showing no hin=
of self-doubt.

Rehbock whips out a small screwdriver, removes the case of a nearby PC, and
plugs into it a card that is the culmination of two years of obsessive work=
A computer-generated landscape materializes on the 19-inch screen of a
Toshiba TV, depicting three spheres drifting through the sky in a graceful,
random ballet above a rippling blue lake. One sphere appears chromed; it
reflects the lake, and the lake reflects it.

This is the kind of demo computer-animation studios used to create 15 years
ago, when several weeks of intensive processing on a chain of SGI
workstations was required to render a two-minute sequence. Today, though,
the VM Labs board is doing it in real time. Calculations that used to
require several hours are now being done in less than a second. Moreover,
the program producing this imagery is tiny - less than 2 Kbytes in size.

"And it uses ray tracing," says Rehbock. "No polygons."


The VM Labs chip can do it, though. Unlike even the newest game systems, it
was purpose-built. "The Sony PlayStation uses a processor similar to the ol=
SGI workstations," says Rehbock. "It's eight-year-old Unix technology.
Nintendo 64 is an R4300i with an SGI reality engine strapped on. The Sega
Saturn uses a Hitachi SH2, which is a processor used in washing machines."
He shrugs dismissively.
But game graphics are just the beginning. The chip can create special
effects such as still frames, slow motion, and scaled and tiled images; or
it can transform the humble DVD player into a videophone, a Web browser, or=
yes, a game machine.
In the past, game systems have been exiled to the kids' room while the
adults watch movies on the TV in the living room. But if a DVD player can d=
games as well as movies, programmers may be tempted to use the extra
features, and the artificial separation of games and "serious" entertainmen=
could break down.
If we want scrupulous realism, this may require even more processing power;
but the VM Labs platform is designed to be upgradable. "We have a processor
road map," says Rehbock, "that takes us from 1,500 millions of instructions
per second today to 24,000 mips around the year 2000. Our goal eventually i=
to give game programmers the ability to do Toy Story in real time, so that
you can interact with it."

After Rehbock finishes his demo, three other men enter the room. They look
uneasy, self-conscious, like woodland creatures emerging from hiding places
and blinking in the sudden light of day. Clearly, these guys are not
accustomed to giving interviews. But they've achieved some wild ambitions
during the past couple of years, and they're well aware of this fact.

Matthew Halfant is a bearded mathematician, a former employee of IBM and
later Apple, who developed low-level routines for the VM Labs chip before
the chip even existed. "I made sure the square roots and reciprocals would
perform well," he says with a funny smile.

At Apple, he worked on advanced font technology. "But the stuff that I did
never saw its way out the door. Another mathematician, whom I know from my
MIT days, got very enthusiastic about some of my work and is using it at
NASA to stabilize helicopters." Halfant adjusts his glasses. "Personally, I
was trying to stabilize fonts."

At VM Labs his work has a better chance of finding its intended market.
Next to Halfant is Louis Cardillo, vice president of engineering. His
hardware-design team consists of just six people, including himself, five o=
them formerly at 3DO - the last great failed attempt to revolutionize

"It's a tiny hardware team to design a chip of this size and complexity,"
says Cardillo. "Many companies would have 30 or 40 people. When Intel gets
around to this problem - as they will, sooner or later - maybe they'll thro=
a thousand people at it.
"But," he smiles, "you get more enthusiastic if there are fewer people,
because you have a bigger chunk of the design. You work harder."


This will be possible because VM Labs doesn't plan to make money by selling
silicon. Its revenues will come from another source entirely, through a
series of unique licensing agreements.

Nicholas Lefevre, chief counsel for the company, spells it out. "Our basic
business model is two-tiered," he says. "First we license semiconductor
companies to make the chip we developed - but they can only sell it to
consumer-electronics companies that also have a license agreement with us.
This is the same system that Dolby uses."
Thus, a famous chipmaker acquires the right to fabricate silicon based on
the VM Labs design, and since it offers extra functionality for negligible
extra cost, several major consumer-electronics companies have already agree=
to buy it. In fact, some will be using it in their entire DVD product lines=

"But the real money will come later, from the second tier," Lefevre adds.
"Any software-development company that wants to write for the new chip will
pay VM Labs a royalty on every copy of each game sold."

It's a classic Silicon Valley vision: VM Labs doesn't build anything - it
licenses its intellectual property to other, bigger firms that do the
capital-intensive work of manufacturing and selling things.

Lefevre has an interesting background. "I went to law school at Georgetown
in the mid-1970s," he says, "but got the computer bug and started staying u=
all night programming Radio Shack's TRS-80, then the Atari 400 and 800."

He tried to set up a private practice in Washington, DC, specializing in
computer-related law, but it wasn't the right place or time, and he became
general counsel for Commodore, in the Commodore 64 era. In 1983 he had the
fine distinction of bringing the first antitrust lawsuit against Microsoft,
when it had an alliance with IBM, which Lefevre termed a "symbiotic

"We asked for $32 million," he recalls. "We met to settle, Gates yelled his
head off, then we settled."

Richard Miller's office is small and unobtrusive, halfway down a hallway at
the back of the building. Miller, likewise, is soft-spoken and
unpretentious. With a pale complexion, intent, serious eyes, and bushy blac=
hair flanking a high rounded forehead, he looks like the stereotypical
scientist who'd rather be in a lab somewhere, wearing a white coat.

His ambitions, though, have always been grand. "When Jack Tramiel
interviewed me for a job at Atari in 1989, he asked what I'd like to be
doing in 5 to 10 years, and I said that within that time I intended to
compete with him." Miller shrugs. "He gave me the job anyway, as vice
president of technology, in charge of all the engineering and R&D."
When Miller left Atari in 1994 with little money and no definite prospects,
he had a clear idea of what he didn't want to do. "I knew I didn't want to
compete in the business of manufacturing and selling consumer electronics,"
he says. "The manufacturing cycle for Christmas begins in June. You have to
anticipate the market beyond any reasonable level of foresight, and it's
essentially impossible. So you often get it wrong, and then you have to
recover. Success is not determined by talent or any other rational basis."

selling it directly. But he still needed capital to take him through the
product-development phase, which entailed a sacrifice he didn't want to
make: surrendering his autonomy to venture capitalists.

Instead, he approached some wealthy friends. "One of them listened to my
plan, which focused on the technology and the opportunity that MPEG-2 was
presenting, and he said, 'Do you have a business model?' And I said, 'Uh,
what's a business model?' He explained that I needed a way to finance
myself, to get to this Valhalla in the future where there would be a
wonderful stream of content royalties." Miller smiles at his own na=EFvet=E9. "=
went home," he says, "and had to think about it."

And he's concerned about potential competitors. "This is no time to rest on
our laurels," he says. "We heard about a company recently that is trying to
do something similar to us. In two years they got through $50 million. In
the first two years of VM Labs we got through $3 million. We've run this on
a shoestring - not entirely by choice."

Still, Miller emphasizes VM Labs's unique advantages. It's well ahead of th=
competition, and he believes that his team is intellectually unmatched.
Also, "we can change direction very quickly, without venture capitalists,"
Miller says. "That's been an absolute pleasure - not worrying whether the
person with the money will agree with our vision. So, I believe we will
consistently beat our competitors with technology, and maintain a lead
there, and maintain a better time to market."

He sounds as if he's rerunning the business plan in his head. In the US,
home movies are a $15 billion industry, home videogames $7 billion, and the
VM Labs chip can handle both, with extra power to spare ... so, yes, it
could work. It should work.

Let's suppose that it does work. What's the best possible outcome that
Richard Miller can imagine?

The question disconcerts him. "We're not great on big fancy visionary quote=
around here," he says with a forced laugh. He pauses for a moment, thinking=
"I hope we will enable software developers to get really creative again," h=
says finally, "and take off the shackles. I hope, and believe, we can
establish a whole new standard for interactivity."

He sounds sincere about this goal, perhaps harking back to the days when he
and his contemporaries were first hooked on computers, long before the era
of huge software companies selling boring products designed by committee an=
slowed by bloated code. Back then, in the late 1970s, one teenager could
create something exciting and wonderful, using fanatical ingenuity to
squeeze it into 48 Kbytes of RAM.

Among the employees at VM Labs, this spirit endures. "They put their lives
into this business," Miller says.

Why do it? In the hope of a big payoff? "No," Miller says. "They just love
working with other great engineers - coming up with ideas, where you know
when you've solved a problem, it's a great solution, and probably no one ha=
ever done it before."

Yet even this is not enough. If a technical tour de force remains a secret
shared only by a few, it has not realized its true potential. It must be
disseminated as widely as possible.

Miller makes no secret that he recognizes this fact. "I'd like our chip to
be the Dolby of interactivity," he says, flatly and simply. "I want it to b=
in every living room in the civilized world."

In other words, total market penetration.

The ultimate Valley dream.

Charles Platt (, a frequent contributor to Wired, wrote
"Breaking the Law of Gravity" in Wired 6.03.