AMD GPU OpenMP 2023

In collaboration with AMD, DiRAC presented a pair of training opportunities in spring 2023 designed to help our users take advantage of some of AMD’s cutting-edge  GPU hardware and software toolstack.

First, AMD experts presented a 2-day online pre-training session that took our users through the basics of how to work with AMD GPUs, including Getting Started with OpenMP and Offloading, and demonstrated some of AMD’s key features and tools.

The pre-training was followed a few weeks later by a 3-day in-person hackathon, hosted at University College London. This event gave our more experienced users and those who took part in the pre-training the opportunity to apply new techniques to their own codes in the company of AMD experts and members of DiRAC’s research software engineering team. Participants were also able to use AMD’s dedicated development cluster to run their codes.

Whilst primarily focussed on practical hands-on application, the hackathon also had a small lecture component, again presented by AMD,  looking at Communication Fabrics, GPU-Aware MPI, and a brief Introduction to Rocprof.  During the event, there were also feedback and Q&A sessions to help spread good practice and address any issues and on the afternoon of the 3rd day participants gave presentations to the group demonstrating what they had achieved.

The event attracted 18 participants from 7 research teams across the DiRAC and wider community, with one additional team being supported remotely. The participants were at all career stages, with RSE’s and academic staff joining students, support staff, and postdocs.  There was a formal three-course dinner sponsored by AMD on the second night which was very well attended and gave a great opportunity for the group to not only build links with other research teams but also with DiRAC and AMD’s technical teams.