Webinar: From traces to optimisation with NVIDIA Nsight Systems on heterogeneous HPC platforms
About the talk:
Modern HPC applications distribute their work across CPU and GPU, and understanding where time is actually spent requires observing both architectures simultaneously within a single coherent framework. This webinar presented NVIDIA NSight Systems as the foundation of performance engineering workflow for NVIDIA architectures: participants learn how to instrument and collect traces with nsys, use NVTX ranges to improve timeline readability, filter the profiling session to the relevant portions of the application, and how to interpret the timeline and post-processing summaries to extract meaningful insight from the data. The session covers the most relevant performance factors in heterogeneous programming, from data locality and stream concurrency to GPU awareness in distributed MPI runs, with the goal of making informed and targeted optimisation decisions.
This webinar was held 10 July 2026.
About the speaker:
Originally trained as a physicist, Laura Bellentani specialised in High-Performance Computing during her PhD in Physics and Nanosciences at the University of Modena and Reggio Emilia. Since 2021, she has been a member of the High-Level Support Team at CINECA, where her work focuses on porting applications to GPUs. She maintains a focus on balancing performance with code maintainability and portability, viewing profiling as an indispensable tool for achieving effective and sustainable optimisations.
About the moderator:
Isabel Rio-Torto received the Ph.D. in Computer Science from the Faculty of Sciences of the University of Porto (FCUP) in 2026 and the master’s degree in Electrical and Computers Engineering from the Faculty of Engineering of the University of Porto (FEUP) in 2019. Alongside her role as a researcher at INESC TEC, she is an Invited Teaching Assistant at FEUP and FCUP, teaching C programming, Computer Vision, and Artificial Intelligence. Isabel has extensive experience using HPC systems, particularly for GPU-based AI training and inference in PyTorch, including work with Large (Vision) Language Models.