In a clear move that demonstrates the flexibility of the PowerPC architecture, IBM is creating new supercomputer solutions for many uses.
The same supercomputing technology IBM is selling to Lawrence Livermore National Labs to model nuclear weapons may also be used to simulate something equally complex, as far as Hollywood animators are concerned: a room full of extras.
IBM is now one year away from building the 65,000-node Blue Gene/L machine for Lawrence Livermore and two working prototypes of the machine have been now been ranked among the ten fastest computers in the world, according to the Top500 list of the world's fastest supercomputers.
With the underlying processor and system design for Blue Gene now essentially completed, IBM's attention is turning to software, according to the company, and in the last few months IBM has begun talking to users and software developers in a variety of industries to see where else Blue Gene's unique design may apply.
"We're asking ourselves new types of questions: Is Blue Gene capable of doing animation? Is Blue Gene capable of doing fluid flow analysis? You pick the topic," said Dave Turek, vice president of Deep Computing at IBM.
Already, Threshold Digital Research Labs, a digital animation company based in Santa Monica, California, has expressed interest.
"We're very, very interested in anything that increases the amount of horsepower we can put on a particular project," said George Johnsen, Threshold's chief animation and technical officer.
Because there is a waiting list to get access to the Blue Gene systems, Threshold has not yet begun work with IBM, but the National Center for Atmospheric Research has, and IBM is now fielding questions from customers in the petroleum, life sciences, and automotive industry as well, according to Alan Gara, chief architect of Blue Gene.
"We have tremendous interest from the outside world to try and get access to the machine," Gara said. "If we really opened up the machine and said, 'Go ahead, everyone can have access to it,' we would be getting many more people. We're kind of throttling them back just because we can't handle them all right now."
Gara's team is hoping that their Blue Gene prototype will surpass Japan's 36-teraflop Earth Simulator – which has been ranked the world's fastest computer since 2002 – by the time the next Top500 list is published next November. A teraflop is one trillion mathematical operations per second.
By early 2005, when Lawrence Livermore's $100 million Blue Gene system is up and running, it should easily eclipse the Earth Simulator with an estimated maximum performance of more than 200 teraflops, Gara said.
Blue Gene uses specially manufactured 700MHz PowerPC processors that are built with what IBM calls a "system on chip" design that is intended to consume minimal power. Blue Gene systems themselves are built into special racks that are tilted to maximize cooling efficiency.
The Lawrence Livermore system is expected to use between 1.6 and 1.7 megawatts of power, far less than the 4.5 megawatts that Livermore expects a 100-teraflop supercomputer called ASCI Purple to use when it is completed in December of this year, said Michael McCoy, deputy associate director for high performance computing at Lawrence Livermore.
"That's a factor of three, and when you figure that you're paying about $1 million per megawatt-year, that makes people like me sit up and listen," he said.
Blue Gene's raw performance could have a similar effect on digital studios, which are constantly striving to outdo each other with animated effects, said Threshold's Johnsen. While studios may now be able conjure realistic-looking characters, such as the computer-generated Gollum in the Lord of the Rings movies, animating something that looks like a living human is a much more difficult.
"You know what the hard part is?" Johnsen asked. "A room full of regular folks. That's a hard challenge. I can make Gollums all day. Everybody can. I can give you dinosaurs, I can give you talking cars. All of that stuff is relatively easy with the current technology. But if I want to give you a human, that's a hard job. The audience is very sophisticated. They know what humans look like."