Performance Portability for Next-Generation Heterogeneous Systems

Perspectives and challenges of future HPC installations for atomistic and molecular simulations

Date: 20 February 2024 at 11:20

Location: Zuse Institute Berlin, Germany

There is a huge and growing diversity in the computer processors used to power the leading HPC systems, AI Research Resources, and supercomputers, and our simulation codes now need to run on CPUs, GPUs, and perhaps other devices such as AI/ML accelerators. With multiple hardware vendors now offering competitive processors, we as a HPC community have more choice, but how do we write our applications to make the most of this opportunity? Our applications need to address the “Three Ps” and be Performance Portable and Productive.

In this talk I will discuss the opportunities this variety of brings to applications and share best practices for rigorously measuring application performance portability and how to compare across diverse architectures. Finally, I will share strategies for writing performance portable programs and present the roles the ISO languages C++ and Fortran and parallel programming models and abstractions such as SYCL, OpenMP and Kokkos might play in the shifting heterogeneous landscape.