P’Con – Getting ready for AMD GPU and Programming Techniques for Exa-scale panel discussion
12:30pm - 2:30pm
As part of the first PaCER Conference – P’con: Setting the pace for exascale – George Markomanolis, Lead HPC Scientist at CSC Finland, supporting Lumi Supercomputing, will talk about CSC’s experience of preparing for AMD GPUs.
George Markomanolis is a Lead HPC Scientist at CSC – IT For Science Ltd in Finland. George is a member of the HPC Support team helping users optimize and port applications to CSC infrastructure. He is part of the team preparing and training users for the LUMI supercomputer, which includes porting codes and training users on how to port codes to AMD GPUs. George has also worked as an HPC Engineer at Oak Ridge National Laboratory. He supported users of Summit as well as other available supercomputing facilities at Oak Ridge Leadership Computing Facility. George contributed to achieving the number 1 ranking in the IO-500 list during Supercomputing 2018 by tuning the benchmark on the Spectrum Scale file system. He is a member of the IO-500 committee.
George’s presentation will be followed by a panel discussion, where he will be joined by Ugo Varetto, Pawsey CTO, and Jordi Blasco, HPCNow! CTO. The panel will discuss Programming Techniques for Exa-scale.
While P’con runs for a week, we will open the conference to the entire HPC and research community to participate in Thursday’s events.
Register to take the opportunity to learn from internationally recognised experts about working with GPU technologies.
Speaker: Dr George Markomanolis
Title: Towards the AMD GPU era
LUMI is a new upcoming EuroHPC pre-exascale supercomputer with a peak performance of a bit over half exaflop by HPE Cray. It is known that this system will be based on the next AMD MI200 GPUs, similar to Setonix’s architecture, and this is a new environment for all of us. In this presentation, we discuss the AMD ecosystem, we present briefly the procedure to convert CUDA codes to HIP, options for codes with OpenACC/OpenMP among also how to port Fortran codes. We discuss other programming models, such as Kokkos, SYCL and we compare some benchmarks across AMD MI100 and NVIDIA V100/A100. Finally, we discuss the challenges and the future directions that a scientist should consider.
Panel Discussion: Programming Techniques for Exascale
George Markomanolis is a Lead HPC Scientist at CSC – IT Center for Science Ltd. His research interests are on performance evaluation/optimization of HPC applications and parallel I/O analysis on various filesystems among porting applications to new technologies. He is also a member of the IO500 committee. In his current role explores programming models for the LUMI supercomputer among other activities. He is an expert on using many profiling tools, understanding the bottlenecks and proposing solutions. Before joining CSC, he was HPC Engineer at Oak Ridge National Laboratory, Computational Scientist at KAUST, and a senior engineer at Barcelona Supercomputing Center. He obtained his M.Sc. in Computational Science from the Department of Informatics and Telecommunication, National and Kapodistrian University of Athens, Greece in 2008 and his PhD in Computer Science from the Ecole Normale Supérieure de Lyon, France in 2014.
Jordi Blasco is a computational physicist from the University of Barcelona, Spain. Jordi has 17+ years of experience in High-Performance Computing (HPC) in industry and academic environments, 12+ years of experience as a solutions architect on mission-critical environments, and 12+ years of experience in leadership and coordination of cutting-edge HPC projects. Jordi has a solid background in parallel programming, performance analysis, application tuning, and HPC system administration. Jordi is a very active contributor to several open-source projects dealing with HPC and large-scale systems, and he has worked as an independent advisor for several companies and research centres. Jordi is the founder of the popular HPC Knowledge Portal (hpckp.org) and co-founder of HPCNow! (hpcnow.com), a global consulting company focused on HPC.
Ugo Varetto, the Chief Technology Officer of the Pawsey Supercomputing Reserch Centre, is an accomplished technology leader and computational scientist. His focus is on data-intensive HPC, large scale parallel I/O, accelerators and quantum technology. He has held key roles in significant research projects, such as the EU-funded ‘Human Brain Project’, has contributed to numerous projects with the European supercomputing infrastructure and has worked in commercial companies and government and academic institutions in the USA, Europe and Australia. Ugo has been developing GPU-accelerated scientific software since 2006 and is currently working on designing scalable scientific computing solutions on hybrid CPU/GPU systems and implementing large-scale, real-time data processing and visualisation workflows, while supporting international research projects such as the Square Kilometre Array (SKA). He is also involved in quantum technology research projects, including the acquisition and operation of a diamond quantum computer at the Pawsey Supercomputing Research Centre.
Registration is required