Nov 09, 2012
Peter Mendygral, Cray / University of Minnesota
"The Future of Computational Astrophysics"
High Performance Computing (HPC) is critical to theoretical and observational discovery in astrophysics. For decades, advances in computing technologies have enabled the community to produce increasingly realistic simulations of stellar atmospheres, supernovae, galaxy evolution and structure formation as only a few examples. Progress like this requires close partnerships between code developers and system designers so that hardware can be designed to meet scientific needs and applications can be designed to realize hardware capabilities. Today's HPC systems are built with multiple levels of parallelism and even heterogeneous resources. Application developers must consider the performance and reliability of their numerical methods on multiple cores in CPUs, accelerator technologies like GPUs, high-speed networks, and storage systems. In this talk I will review the state of the HPC industry and its impact on computational astrophysics today. I will also discuss the trajectory of the industry and what types of astrophysical applications, simulations and analysis might be possible as we enter the era of Exascale.