Profiler Training Series, Part 2: Advanced, How to Profile

Event Details

Date and Time:
Thursday, February 12, 2026 from 1:00 PM to 2:30 PM Eastern Time

Presenters:
AMD Performance Tools Team in collaboration with Oak Ridge Leadership Computing Facility staff

About the Series:
This session is part of the Profiler Training Series, a multi-part training designed to help users build practical performance profiling skills on Frontier. View the full series schedule and upcoming sessions on the Profiler Training Series page.

Hackathon Connection:
Participation in this training series is strongly encouraged for teams planning to apply to the 2026 Spring Frontier Hackathon. The skills covered will help teams prepare competitive abstracts and optimize applications ahead of the hackathon. Visit the Frontier Hackathon page for event details and application information.

Event Description

This session continues from Part One of the How to Profile series and builds on the foundational skills learned in basic performance profiling. Participants will learn how to systematically analyze and optimize multi-process GPU applications that use the Message Passing Interface (MPI).This workshop provides a deeper understanding of how to connect GPU and network performance insights into a unified view of your application’s behavior, enabling more effective optimization on large-scale systems.

What You Will Learn

• Multi-Process Profiling Basics:How to profile an MPI-based multi-process job and measure where your application spends time.
• Network Profiling and Optimization: How to analyze MPI communication patterns, understand inter-node data exchange, and identify bottlenecks at scale.
• Advanced GPU Kernel Optimization: How to apply insights from rocprof-computeand rocprof-sys to deeply understand GPU kernel performance and bridge performance gaps across systems.

By the end of this session, you will be able to identify optimization opportunities across GPU computation and MPI communication layers—empowering you to tune your multi-process application for AMD systems such as Frontier.

References

• Performance Profiling on AMD GPUs – Advanced Usage

Who Can Attend
Anyone in the HPC community is welcome to join. Those planning to submit applications for the 2026 Frontier Hackathon (Due on March 4) are highly encouraged to attend. Git repositories with the examples covered are provided, but participants will not be expected to run them during the session.