ORNL Management Talks Titan at International Supercomputing Conference
Five-day gathering in Hamburg, Germany focuses on next-generation computing issues
The 2012 International Supercomputing Conference (ISC), which took place June 17–21 in Hamburg, Germany, brought together leading high-performance computing (HPC) experts for presentations, talks, and exhibitions to move the world’s fastest machines into the next era.
ORNL sent a contingent of leading computational experts to explain how America’s largest computing facility currently operates and to detail its plans for the future.
“ISC provides a unique opportunity to easily engage with European supercomputing facilities, allowing us to share best practices and innovative approaches to solving challenges often unique to HPC,” said Galen Shipman, data systems architect in the computing and computational sciences directorate at the OLCF.
Shipman gave three talks at the conference addressing one of the largest problems facing supercomputers—data management. As datasets for computer simulations grow exponentially larger, computer scientists must find new ways of storing and maneuvering data on and off of high-performance machines.
Both Buddy Bland, project director for the OLCF, and Jack Wells, director of science for ORNL’s National Center for Computational Sciences, gave multiple presentations during the conference. Bland and Wells discussed how the OLCF is converting their Cray XT5 Jaguar supercomputer—currently the sixth fastest supercomputer in the world—into Titan, a Cray XE6 supercomputer that will be capable of approximately 20 petaflops, or 20 quadrillion calculations per second, when fully operational in early 2013.
Titan will have a hybrid computing architecture using both high-speed graphics processing units (GPUs) and traditional central processing units (CPUs) to make for a faster and more efficient machine. A ten-cabinet slice of Jaguar has been modified with NVIDIA GPUs to allow researchers to adapt to the novel computing environment. Both Bland and Wells presented at NVIDIA’s satellite event, highlighting how GPUs will push science forward.
“Hybrid architectures are critical if we are to build faster computers and live within the power constraints of our supercomputer centers,” said Bland. “Titan will have a mix of processors designed to provide fast performance on serial tasks, as well as processors designed to efficiently processes parallel tasks.”
Julia White, manager of the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program, spoke about how to access supercomputing resources. The INCITE program, funded by the Department of Energy and jointly managed by the Oak Ridge and Argonne Leadership Computing Facilities, provides competitive access to researchers trying to solve grand-scale problems in a variety of scientific disciplines. —by Eric Gedenk