APPLICATIONS
New Cray X1 at Oak Ridge is up to 25 Times Faster on Scientific Applications
- Written by: Writer
- Category: APPLICATIONS
High-End Computer Is 50 Percent Faster Per Processor Than Japan's Earth Simulator On Standard Climate Problem -- Global supercomputer leader Cray Inc. (Nasdaq NM:CRAY) today reported that the new Cray X1(TM) supercomputer is running challenging applications up to 25 times faster than previously achieved by the Department of Energy's Center for Computational Sciences (CCS) at the Oak Ridge National Laboratory (ORNL). In the ORNL evaluations, the new high-end computer also has run a standard climate modeling application 50 percent faster per processor than Japan's Earth Simulator, which currently is recognized as the world's fastest supercomputer and was designed primarily to run climate modeling problems. The U.S. Department of Energy's Oak Ridge National Laboratory (ORNL) was selected to test the effectiveness of the new Cray X1 supercomputer, and ORNL's Center for Computational Sciences now houses a 256-processor Cray X1 system. ORNL Director Jeff Wadsworth said new funding by the DOEs Office of Advanced Scientific Computing Research (OASCR) and the Scientific Discovery Through Advanced Computing (SciDAC) program has enabled significant recent advances in several application areas: -- Los Alamos National Laboratory's widely used parallel ocean program (POP v1.4.3) has already achieved a 50 percent performance improvement on the Cray X1 system, on a processor-by-processor basis, compared to Japan's Earth Simulator. -- Large-scale simulations of high-temperature superconductors ran 25 times faster than previously achieved. -- An important physics application, global GYRO transport calculations, ran on the Cray X1 system 16 times faster, allowing simulations to be run for twice as long as before, and fully resolving the questions raised by a prior study. -- Molecular dynamics simulations related to the phenomenon of photon echo ran on the new Cray X1 system eight times faster than previously achieved. "The early speed and efficiency of the Cray X1 are exciting," said Thomas Zacharia, ORNL associate director in charge of computational sciences. "On the Cray X1, we are regularly seeing efficiencies of 30-35 percent on our applications, and in some cases up to 60-70 percent. We need to see how the Cray X1 does with thousands of processors, and hope we will be able to deploy a machine that large to advance simulation as a peer to theory and experiment. Indications so far are that a Cray X1 supercomputer of sufficient size could match or exceed anything on earth." "Even at 256 and fewer processors, the Cray X1 system is already outperforming other U.S. supercomputers with thousands of processors, and is enabling leading researchers to advance the boundaries of science and engineering on real-world applications," said Cray Chairman and CEO Jim Rottsolk. "We believe the Cray X1 supercomputer's lead will increase substantially in larger system sizes." Rottsolk said the Cray X1 product reflects the company's strategy to deliver high-efficiency, high bandwidth supercomputer systems.