Many of today's high-end HPC systems are equipped with GPUs. This is true for smaller clusters, Tier-0 and Pre-Exascale systems alike. Using modern GPUs can significantly improve the computational performance achieved. Making use of these resources will be even more crucial in the future as the trend towards GPUs will in all likelihood continue. GPUs are currently becoming mainstream within the Exascale Computing Project in the US. In this PRACE Autumn School participants were provided with the necessary skills and tools to efficiently use GPUs in their daily work.
The school addressed participants from all scientific fields. They were taught the necessary programming paradigms and were guided to write correct and efficient code for GPUs, including learning how to port existing code to use GPU acceleration. The course included many hands-on exercises so that the learned concepts can be applied immediately. In addition, several invited speakers with different backgrounds reported on successfully porting their scientific applications to GPUs.
Case studies and hands-on tutorials form an integral part of the school. The computational resources necessary to conduct the exercises were provided in form of a GPU cluster at the University of Innsbruck, which was reserved for exclusive usage for the participants of the school. This was augmented by a second batch of GPUs from VSC-3, a cluster located at the national centre of supercomputing in Austria, with likewise exclusive reservation for the entire period of the school.
A PRACE event, participation was free of charge for over 50 participants from European or PRACE member countries.
Link to PRACE event page: https://events.prace-ri.eu/event/1154/