Public/Private Collaboration Aims Toward Exascale With Smaller, Faster, Sustainable Machines Based on Heterogeneous Computing

ATLANTA - October 21 2009 - The Georgia Institute of Technology todayannounced its receipt of a five-year, $12 million Track 2 award fromthe National Science Foundation's (NSF) Office of Cyberinfrastructureto lead a partnership of academic, industry and government experts inthe development and deployment of an innovative and experimentalhigh-performance computing (HPC) system. The award provides for thecreation of two heterogeneous, HPC systems that will expand the rangeof research projects that scientists and engineers can tackle,including computational biology, combustion, materials science, andmassive visual analytics. The project brings together leading expertiseand technology resources from Georgia Tech's College of Computing, OakRidge National Laboratory (ORNL), University of Tennessee, NationalInstitute for Computational Sciences, HP and NVIDIA.

NSF's Track 2 program is an activity designed to fund the deploymentand operation of several leading-edge computing systems operating at ornear the petascale. An underlying goal is to advance U.S. computingcapability in order to support computational scientists and engineersin the pursuit of scientific discovery. The award announced today isthe part of the fourth round of awards in the Track 2 program.

"Our goal is to develop and deploy a novel, next-generation systemfor the computational science community that demonstrates unprecedentedperformance on computational science and data-intensive applications,while also addressing the new challenges of energy-efficiency," saidJeffrey Vetter, joint professor of computational science andengineering at Georgia Tech and Oak Ridge National Laboratory.

"The user community is very excited about this strategy," Vettercontinued. For example, James Phillips, senior research programmer atthe University of Illinois who leads development of the widely-usedNAMD application, says "Our experiences with graphics processors overthe past two years have been very positive and we can't wait to explorethe new Fermi architecture; this new NSF resource will provide an idealplatform for our large biomolecular simulations."

Georgia Tech's Vetter will lead the five-year project as principalinvestigator. The project team is comprised of luminaries in the HPCfield, including a Gordon Bell Prize winner and previous recipients ofthe NSF Track 2B award. Co-principal investigators on the project areProf. Jack Dongarra (University of Tennessee and ORNL), Prof. KarstenSchwan (Georgia Tech), Prof. Richard Fujimoto (Georgia Tech), and Prof.Thomas Schulthess (Swiss National Supercomputing Centre and ORNL).

The platforms will be developed and deployed in two phases, withinitial system delivery planned for deployment in early 2010. Thissystem's innovations in performance and power will be achieved throughheterogeneous processing based on widely-available NVIDIA graphicsprocessing units (GPUs). As industry partners, HP and NVIDIA will beproviding the computational systems, platforms and processors needed todevelop the system.

"Research institutions are looking for energy-efficient,high-performance computing architectures that can speed time tosolution," said Ed Turkel, manager of business development in theScalable Computing and Infrastructure business unit at HP. "Thecombination of HP's industry-standard HPC server technology with NVIDIAprocessors delivers increased performance and faster applicationdevelopment, accelerating higher education research projects."

The initial system will pair hundreds of HP high-performance Intelprocessors with NVIDIA's new next-generation CUDA architecture,codenamed Fermi, designed specifically for high-performance computing.This project will be the first of the Track 2 awards to realize thevast potential of GPUs for HPC.

"Computational science is a key area driving the worldwideapplication of GPUs for high-performance computing," said Bill Dally,chief scientist at NVIDIA. "GPUs working in concert with CPUs is thearchitecture of choice for future demanding applications."

A critical component of the program is a focus on education,outreach and training to expand the knowledge and understanding of HPCamong a broader audience. The Georgia Tech team will conduct workshopsto attract and train new users for the system, engage historicallyunderrepresented groups such as women and minorities, and educatefuture generations on the vast potential of high-performance computingas a career field.

More information on the project and its resources is available at http://keeneland.gatech.edu.

About the Georgia Institute of Technology

The Georgia Institute of Technology is one of the nation's premierresearch universities. Ranked seventh among U.S. News & WorldReport's top public universities, Georgia Tech's more than 19,000students are enrolled in its Colleges of Architecture, Computing,Engineering, Liberal Arts, Management and Sciences. Tech is among thenation's top producers of women and African-American engineers. TheInstitute offers research opportunities to both undergraduate andgraduate students and is home to more than 100 interdisciplinary unitsplus the Georgia Tech Research Institute

For more information, contact:
Stefany Wilson
Georgia Tech College of Computing
404.894.7253
stefany@cc.gatech.edu