1 / 13

Technology Challenges for Scientific Computing

Technology Challenges for Scientific Computing. Dan Hitchcock Advanced Scientific Computing Research Daniel.Hitchcock@science.doe.gov. Overview. Evolution of semiconductors and microprocessors will place significant challenges on all of scientific computing;

elwyn
Download Presentation

Technology Challenges for Scientific Computing

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Technology Challenges for Scientific Computing Dan Hitchcock Advanced Scientific Computing Research Daniel.Hitchcock@science.doe.gov

  2. Overview • Evolution of semiconductors and microprocessors will place significant challenges on all of scientific computing; • Decreases in clock speed to save power • Increase in number of simpler cores on chips • Memory – cpu communication • Power constraints • End of constant memory/process weak scaling • Synchronization • These problems are most apparent in moving to exascale but impact computing at all levels. DH NCSA April 20, 2010

  3. Identifying exascale applications and technology for DOE missions • Town Hall Meetings April-June 2007 • Scientific Grand Challenges Workshops November 2008 – October 2009 • Climate Science (11/08), • High Energy Physics (12/08), • Nuclear Physics (1/09), • Fusion Energy (3/09), • Nuclear Energy (5/09), • Biology (8/09), • Material Science and Chemistry (8/09), • National Security (10/09) • Cross-cutting workshops • Architecture and Technology (12/09) • Architecture, Applied Mathematics and Computer Science (2/10) • Meetings with industry (8/09, 11/09) MISSION IMPERATIVES FUNDAMENTAL SCIENCE DOE Exascale Initiative

  4. The Fundamental Issue Intranode/SMP Communication Intranode/MPI Communication On-chip / CMP communication DH NCSA April 20, 2010

  5. Memory TechnologyBandwidth costs power DH NCSA April 20, 2010

  6. Potential System Architectures DH NCSA April 20, 2010

  7. On-Chip Architecture:different approaches to on-chip clustering • Cost of moving long-distances on chip motivates clustering on-chip • 1mm costs ~6pj (today & 2018) • 20mm costs ~120 pj (today & 2018) • FLOP costs ~100pj today • FLOP costs ~25pj in 2018 • Different Architectural Directions • GPU: WARPs of hardware threads clustered around shared register file • CMP: limited area cache-coherence • CMT: hardware multithreading clusters DH NCSA April 20, 2010

  8. More Bad News • I/O to disk will be relatively slower than it is today so traditional checkpointing not practical; • Part of the file system may be on the node; • There will be more silent errors; • Weak scaling approaches (constant memory/flop) probably will not work; • Bulk synchronization very expensive. DH NCSA April 20, 2010

  9. Approaches • Locality, Locality, Locality! • Billion Way Concurrency; • Uncertainty Quantification including hardware variability; • Flops free data movement expensive so: • Remap multiphysics to put as much work per location on same die; • Include embedded UQ to increase concurrency; • Include data analysis if you can for more concurrency • Trigger output to only move important data off machine; • Reformulate to trade flops for memory use. • Lightweight operating systems • What to do about cache coherence DH NCSA April 20, 2010

  10. Possible Machine Abstract Models DH NCSA April 20, 2010

  11. Workshops Past • ExascaleCross-cutting workshops • Architecture and Technology (12/09) • Architecture, Applied Mathematics and Computer Science (2/10) • Heterogeneous Multicore Consortium 1/10 • IESP Oxford April 2010 • Upcoming • Data Management, Analysis & Visualization PI Meeting: Focus on Impact of Exascale Architectures & Preparing for Co-Design Process, Tentatively scheduled for August 17-19, in Santa Fe, NM • Best Practices Workshop on Power September 2010 DH NCSA April 20, 2010

  12. Current Funding Opportunity Announcements • Applied Math • Advancing Uncertainty Quantification (UQ) in Modeling, Simulation and Analysis of Complex Systems -- $3M / year for 3 years to fund 2-6 awards, closes April 26, 2010 • Development of highly scalable approaches for uncertainty analysis in the modeling and simulation of complex natural and engineered systems. • Computer Science • X-Stack Software Research -- $10M / year for 3 years to fund 4-5 awards, closes April 2, 2010 • Development of a scientific software stack that supports extreme scale scientific computing, from operating systems to development environments. • Advanced Architectures and Critical Technologies for Exascale Computing -- $5M / year for 3 years to fund 4-5 awards, Closed March 26, 2010 • Design of energy-efficient, resilient hardware and software architectures and technology for high performance computing systems at exascale. • Scientific Data Management and Analysis at the Extreme Scale -- $5M / year for 3 years for 10-15 awards, Closed March 18, 2010 • Management and analysis of extreme-scale scientific data in the context of petascale computers and/or exascale computers with heterogeneous multi-core architectures. DH NCSA April 20, 2010

  13. Webistes • www.science.doe.gov/ASCR • www.exascale.org DH NCSA April 20, 2010

More Related