sub header2

[View PDF]

 

Kenneth Moreland, SNL

Objective

  • Develop readiness for scientific data analysis and visualization at extreme scale.
    • Address challenges of emerging architectures.
  • In addition to designing our own algorithms, build a toolkit that enables others to build algorithms.

Technology

  • The Dax Toolkit: a visualization toolkit containing a framework that reduces the challenges of writing highly concurrent algorithms.
  • Current investment is 3 year project.
  • Supports simple porting across CPU and GPU architectures.
  • Algorithms written at higher abstraction have performance comparable to alternates written by experts with APIs providing full access to parallel features.

Impact

  • Dax applied to analysis of N-body cosmology simulation to identify void, pancake, filament, and clump features.
  • Requires expensive operation of finding cells in irregular, self-intersecting mesh.
  • Dax demonstrates finding cells while yielding speedups of up to 22× with multiple cores and 65× using a GPU.

Threshold (No Point Merging)

Threshold (With Point Merging)

Speed up compared to serial execution

Speed up compared to perfect linear scaling