updated 09:40 am EDT, Tue October 11, 2011
Tesla GPUs could deliver up to 20 petaflops
The Oak Ridge National Laboratory will use the parallel processing capability of the NVIDIA Tesla M2090 GPU in a new supercomputer, a Cray XK6 nicknamed "Titan." The new supercomputer could deliver over 20 petaflops, making it more than twice as fast as the most powerful supercomputer today, the Fujitsu K supercomputer. Titan will also be three times more energy efficient.
One petaflop equals one quadrillion (ten to the 16th power) floating point operations per second.
The computer will be developed in two phases. The first phase is already underway, and will upgrade Oak Ridge's Cray XT5 "Jaguar" supercomputer with 960 Telsa M2090 GPUs. These use NVIDIA's "Fermi" architecture. The second phase will use up to 18,000 Tesla GPUs with next-generation "Kepler" architecture. The second phase is expected to begin in 2012.
The Tianhe-1a supercomputer used a similar approach, using 7,168 NVIDIA Tesla M2050 GPUs to edge the Jaguar last year as the world's fastest supercomputer. The Tianhe-1a was recognized as the world's fastest computer until the K supercomputer eclipsed it. The Titan supercomputer will be roughly four to eight times faster than the Tianhe-1a.
Jeff Nichols, associate laboratory director for Computing and Computational Sciences at Oak Ridge National Laboratory, said the new supercomputer will be used to research a broad range of fields, including material science, energy technology, medical research, and geoscience. More commercially viable biofuels, cleaner burning engines, safer nuclear power, and more efficient solar power are among the planned investigations for the Titan supercomputer.