Sony PlayStation 3 users are engaged in a project to help understand, prevent and maybe even cure disease. And all they have to do is leave their machines on.
Stanford University-sponsored project Folding@Home signed nearly 15,000 PS3 users over its recent launch weekend. Since then, the users have been cranking computer power to help the project achieve its goal: simulating protein misfolding in an attempt to understand disease.
Proteins self-assemble, or “fold,” into position in the human body. It’s how all of our cells are formed. When they misfold, diseases can result. Associate Professor Vijay Pande, who has run Folding@Home since he launched it in 2000, hopes to use the project to find cures for diseases or to prevent them by understanding how proteins misfold.
Pande welcomed the PS3 horsepower. “The PlayStation 3 is 20 times faster than a PC,” he told internetnews.com. “A factor of 20 is mind-blowing. You couldn’t order a PC from Dell or HP that’s 20 times faster than that on the market now; that doesn’t exist.”
The PS3 team was cranking out 338 teraflops of computing power by the time its participant count reached 13,000, while the PC client group, with 160,000 participants, was producing just 151 teraflops of power.
That teraflop performance number has dropped since, which Pande attributes to people not running their consoles as much. “When they first started, a lot were running it 24/7, but then backed off and just ran it overnight,” he said.
Even so, current stats put the PS3 community at 299 teraflops, faster than the fastest supercomputer in the world, the 280 teraflop Blue Gene/L from IBM, and that monster fills a room at the Lawrence Livermore National Laboratory. One of the tasks Blue Gene/L does is protein-folding simulations, just like Folding@Home.
Pande said that he takes a very conservative approach to calculating performance and expects to break the petaflop barrier.
“Getting to a petaflop will be easy in the future, especially as we move away from more conservative [benchmark] numbers to more traditional numbers. I think a petaflop is very, very much within reach. Based on other people’s accounting, we’re already doing a petaflop. What’s more exciting is reaching two or three petaflops,” he said.
Currently, the work units for Folding@Home on the PS3 take eight hours to process, but Pande is looking to cut that down to more manageable four- and two-hour units.
The increased speed of the PS3, and of hardware in general, means more attempts at simulations. Because simulations take less time to process, Pande and the researchers in his groups can do more tests.
Advances in computing are also helping. Pande said his group can get results in just two weeks that used to take over a year with older systems. He said an XBox 360 client is being considered, but they have no plans for a Nintendo Wii release.
The project has moved beyond just processor-based crunching to using the GPU in high-end video cards. Now Pande is looking at multi-core versions of the Folding@Home client. His programmers have been testing it on eight-core processors and getting tremendous results.
In ways, he said distributed computing for science is a lot like Linux clusters in 1995. “At that time, it was clear that it had great potential, but it was a bleeding edge thing to do. It required a lot of knowledge that most groups didn’t have or didn’t want to have.
“Now, 10 years later, Linux clusters are the backbone of how academic scientific computing is done. I see distributed computing as the natural extension to this. Instead of running on PC’s in your lab, you run on PC’s elsewhere. It will be interesting to see where we are 10 years from now,” he said.
