Training

Learn how to profile and trace applications on AMD CPUs and GPUs using the TAU Performance System® (tau.uoregon.edu). This tutorial will cover performance profiling, memory and I/O analysis, OpenMP offload instrumentation, and visualization with Vampir and ParaProf. Participants will also see how TAU integrates with ROCm tools (rocprofiler-sdk) and the E4S software stack (e4s.io).

The Message Passing Interface (MPI) is an industry standard way for the parts of a parallel program to communicate. Since its inception in the 1990s, the MPI standard has evolved through several major revisions. Recently the cray-mpich/9.0.x module(s) have been made available on the Frontier system, which add support for the MPI-4.0 standard. This presentation will discuss the important new features in HPE Cray MPI that users and developers on Frontier might want to use. For example, there are now routines for persistent collectives, and large-count versions of routines that allow specifying more than two billion items in a single call.

OLCF, in partnership with AMD, HPE, and others, invites research teams to participate in a competitive hackathon on Frontier, the world’s most powerful supercomputer. Participants will work to advance fundamental scientific discovery by exploring strategies to optimize performance, scalability, and data processing, as well as developing innovative methods to accelerate high-performance computing workflows and enhance computational efficiency. Selected teams will collaborate with leading experts to push the boundaries of what is possible on exascale systems. Proposals are due by March 4.