Performance Analysis for new HPC programming models
Thursday, July 1, 2021 1:35 PM to 1:55 PM · 20 min. (Africa/Abidjan)
Stream#1
Information
Contributors:
Abstract:
The HPC hardware architectures and the software requirements do not allow anymore to stick with the established HPC parallel programming models. As most important reason because hardware architectures became too diverse and heterogeneous in nature and the #1 machines from the Top500 list during the last few years changed between quite different architectures. If performance is an important concern -- it is by definition in HPC -- it needs to be measured, evaluated, and optimized. What is the state of software tool support for new HPC programming models and their parallelization approaches? How is the convergence of different GPU programming models reflected and what about Kokkos and HPX?
Abstract:
The HPC hardware architectures and the software requirements do not allow anymore to stick with the established HPC parallel programming models. As most important reason because hardware architectures became too diverse and heterogeneous in nature and the #1 machines from the Top500 list during the last few years changed between quite different architectures. If performance is an important concern -- it is by definition in HPC -- it needs to be measured, evaluated, and optimized. What is the state of software tool support for new HPC programming models and their parallelization approaches? How is the convergence of different GPU programming models reflected and what about Kokkos and HPX?
