1 / 9

Panel on Next-Generation Codes/Portability

Panel on Software for Future Architectures. Panel on Next-Generation Codes/Portability. Hank Childs, LBNL Jeremy Meredith, ORNL Pat McCormick, LANL Chris Sewell, LANL Ken Moreland, SNL. April 25, 2012. Dept. of Energy Computer Graphics Forum 2012. Exascale vis & analysis in a nutshell.

Download Presentation

Panel on Next-Generation Codes/Portability

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. Panel on Software for Future Architectures Panel on Next-Generation Codes/Portability Hank Childs, LBNL Jeremy Meredith, ORNL Pat McCormick, LANL Chris Sewell, LANL Ken Moreland, SNL April 25, 2012 Dept. of Energy Computer Graphics Forum 2012

  2. Exascalevis & analysis in a nutshell • Multiple science communities have declared that ExaFLOP-compute capabilities will revolutionize their field Climate Fusion Biology Nuclear Physics Nuclear Reactors High Energy Physics Material Science & Chemistry National Security

  3. Exascalevis & analysis in a nutshell • Power is the central issue at the exascale • ORNL Jaguar: 2 PFLOPs, 6MW • Exascale machine: 1 EFLOP, 20 MW • Every aspect of the machine and the simulation process must be re-thought to minimize power • Plenty of other challenges too!! (e.g. resiliency, memory constraints, etc.)

  4. Exascalevis & analysis in a nutshell • Power is dictating a thrust towards many-core c/o SciDAC Review 16, February 2010

  5. Exascale vis & analysis in a nutshell • Data movement will be a leading cause of power consumption c/o John Shalf, LBNL

  6. Exascale: a heterogeneous, distributed memory GigaHz KiloCore MegaNode system ~3 c/o P. Beckman, Argonne

  7. Exascalevis & analysis in a nutshell • We must minimize data movement: • This motivates in situ processing • Which means that we likely will be running on many-core architectures, alongside the simulation • And VTK is not many-core capable… • … but technologies are emerging that are many-core capable. Within a node Between nodes To and from disk

  8. Format for this panel • Hank Childs, intro, 10 minutes • Jeremy Meredith, EAVL, 20 minutes • Pat McCormick, Scout, 20 minutes • Chris Sewell, PISTON, 20 minutes • Ken Moreland, DAX, 20 minutes • Discussion: 30 minutes

  9. Questions for the panelists • What fundamental problem are you trying to solve? • What are your plans to deal with exascale-specific issues (massive concurrency, distributed memory, memory overhead, fault tolerance)? • What is your philosophy for dealing with ambiguity of the exascale architecture (multiple swim lanes, heterogeneous architectures)? • How is your technology implemented? • What is the long-term result for this effort? (Production software? Research prototype?)

More Related