A recent U.S. Department of Energy workshop concentrating on exascale challenges and current gaps in research and ideology offered insight for those seeking a disruptive approach to exascale computing.
Massachusetts Institute of Technology professor Anant Agarwal said the problem of boosting performance while using less energy is superseded by the far greater challenges of exascale programmability and resiliency, which require disruptive research to address. Such research will emphasize that solving two of the three challenges of performance, efficiency, and programmability will be relatively easy compared to solving all three.
Meanwhile, NVIDIA's Bill Dally said that "algorithms should be designed to perform more work per unit data movement" and that "programming systems should further optimize this data movement." Dally further noted that architectures must effect data movement by furnishing an exposed hierarchy and efficient communication.
University of California, San Diego professor Allan Snavely discussed locality, and how his team is devising tools and methodologies that can identify location in applications to lower the processor frequency for effective power efficiency.
From HPC Wire
View Full Article
Abstracts Copyright © 2011 Information Inc. , Bethesda, Maryland, USA