It turns out that research labs love gaming hardware for much the same reason Wall Street firms do—raw processing power.
While today’s technology enables traders to process more trades, and earn more money, faster equipment in research facilities is allowing scientists to tackle research that would have been impossible even a few years ago. New Scientist details the exploits of MIT’s Nicolas Pinto, who built a 16-GPU (graphics processing unit) super computer for less than $3,000. Pinto is using the PC as part of his bid “to crack the brain's formula for recognizing objects in images.”
His PC performs “statistical analysis in both space and time on thousands of frames of video to find objects moving through the scene.” Pinto added, “It's very cheap to buy a GPU and explore.”
Oak Ridge National Laboratory in Tennessee is attempting to build the world’s fastest computer and is also relying on GPUs. Associate Lab Director Jeff Nichols told The Register that such a PC, which will reportedly use Nvidia’s Fermi graphics processors, would allow “substantial scientific breakthroughs that would be impossible without the new technology.”
Oak Ridge already has GPU clusters deployed for tasks such as climate modeling and simulating supernovas.