Martin Berzins (Sci Institute, University of Utah)
Multiscale and Multiphysics Computations on Presnt and Future Architectures
The challenge of undertaking solving large and complex multiscale and multiphysics calculations on present and future large scale computer architectures is addressed. The computational framework employed is the Utah Uintah software. This software employs a layered approach of an abstract graph-based formulation that generates tasks for a sophisticated runtime system that executes the tasks asynchronously and adaptively. The success of this approach is briefly described and a particular case study used to describe the solution of a challenging combustion problems. A key feature of this problem is the need to model radiative heat transfer by ray tracing. The implementation of this approach is described and its performance examined on both GPus and CPUs.