Guest Lecture by Ruben Laso
Date, time, and location:
Tuesday, July 1st, 11:15 am in the Humboldt-Kabinett (aka HUK, Rudower Chaussee 25, 1st floor)
Abstract:
In this presentation, we will cover three past and ongoing projects related to High-Performance Computing (HPC).
The first part will focus on NUMA (Non-Uniform Memory Access) systems, showing how process and memory placement can be improved to reduce execution times.
The second part will discuss the feasibility of using standard C++ for performance-portable HPC applications, showing how C++ can be used to write code that runs efficiently on different hardware architectures without sacrificing performance.
Finally, we will explore the optimisation of GPU-GPU communication to improve the performance of AI applications. As modern workloads often involve significant data movement, there is an opportunity to tune communication libraries for better performance.
Bio:
Ruben Laso is currently a postdoctoral researcher at the Research Group for Scientific Computing at the University of Vienna. He holds a PhD in High-Performance Computing (2023), a Master’s in Industrial Mathematics (2019), and a Bachelor’s in Computer Science (2017), all from the Universidade de Santiago de Compostela. His research interests include parallel computing, with a particular focus on manycore and NUMA systems, as well as performance portability in scientific codes.
Ruben will be visiting FONDA until mid-July.