"Accelerating Your Applications with AMD GPUs" course: a recap

The course provided a comprehensive introduction to the AMD Instinct GPU architecture and its ROCm ecosystem. Participants learned about various tools and programming models essential for developing or porting HPC or AI applications to AMD GPUs. The focus was on the MI200 and MI300 series accelerators, covering a wide range of GPU programming models, including OpenMP, HIP, and Kokkos. Participants had access to the AMD Accelerator Cloud, which featured MI300, and ARCHER2, which features MI210. 

Day-by-day breakdown 

The first day kicked off with an introduction to the AMD programming model for the MI200 and MI300 series, followed by hands-on exercises. Participants were then introduced to OpenMP offloading, with sessions covering real-world language constructs and advanced OpenMP techniques like zero-copy, debugging, and optimization. 

Day two focused on HIP and ROCm, with detailed sessions on porting code to HIP and optimizing HIP code. The day concluded with a session on the interoperability between OpenMP and HIP, providing participants with practical exercises to reinforce their learning. 

Day three delved into performance optimisation techniques for AMD GPUs. Participants learned about GPU-aware MPI and how to leverage it for better performance scaling. 

The final day introduced participants to the AMD debugging and profiling tools, including 'rocgdb', `rocprof`, 'omnitrace', and 'omniperf'. 

Participant feedback 

The course feedback was overwhelmingly positive. Participants appreciated the comprehensive selection of topics, which were delivered by knowledgeable experts from around the world. The course was praised for its excellent organisation, high-quality materials, and responsive support. Attendees found the breadth and depth of approaches and tools covered to be very beneficial, with exercises that reinforced learning. The presentations were clear and relevant, making complex topics easy to understand. Overall, the course was seen as highly valuable and well-executed. 

Conclusion 

Overall, the “Accelerating Your Applications with AMD GPUs” course was a resounding success. It equipped participants with the knowledge and skills needed to harness the power of AMD GPUs for their HPC and AI applications. The AMD team did an excellent job of delivering this comprehensive training, ensuring that all participants left with a deeper understanding of GPU programming and optimisation. 

Further information

ARCHER2 is the UK's national supercomputing service. This world-class advanced computing resource for UK researchers is operated by EPCC and hosted at the Advanced Computing Facility.

EPCC also provides training for users of the ARCHER2 service. While these courses focus primarily on the ARCHER2 system, they also include courses of wider interest, including courses on MPI and OpenMP. Many course videos and course materials are available. Keep an eye on the ARCHER2 training page for future opportunities to enhance your skills and knowledge in high-performance computing. 

Recordings of the course "Accelerating Your Applications with AMD GPUs"  are available on the ARCHER2 YouTube Channel.