GPUs contributing petaflops to science research
updated 03:30 pm EDT, Tue August 26, 2008
GPU supercomputing
Stanford's Folding@home (FAH) distributed computing project, with the help of GPUs, has produced information contained in over 50 peer-reviewed published scientific papers. The scientists are using the idle power of millions of processors from users all over the world to help study protein folding. A free program can be downloaded that will use an idle computer or Playstation 3 to run scientific algorithms and communicate the results to the main project.
The statistics showed that GPUs were much faster with the scientific tasks. Over 3000 teraflops of processing power come from GPUs and Playstation 3's Cell processors, which account for 90 percent of the project's numbers overall. This same group accounted for less than twenty percent of the total number of active processors being used.
The FAH team developed specialized clients to bring out the individual capabilities of each system. For GPU's such as NVIDIA's Geforce, the researchers built clients based on the CUDA programming architecture to utilize the potential of multi-core parallel processing.
The data is used to develop the understanding of how proteins act in the human body. Some of the information is helping understand more about diseases thought to be related to "misfolding" of proteins such as Alzhiemers or Mad Cow Disease. Software needed to become a contributor is available from Folding@home.
