April 30, 2025 - by CSCS
Eurohack25 is a hackathon devoted to porting scientific applications to GPUs or other massive parallel architectures, as well as optimizing existing high-performance computing applications. A special focus will be on artificial intelligence applications.
High Performance Computing utilizes more and more parallelism provided by modern supercomputers. One development is General-Purpose Graphics Processing Units (GPGPUs), which offers exceptionally high memory bandwidth and performance for a wide range of applications together with many parallel programming units. Another development is more parallelism also for multicore processors. Today, these devices can be programmed with the CUDA/C++ programming platform, HIP or with OpenACC Directives for Accelerators, which offer straightforward extensions to C++ and Fortran to address this programming hurdle. Alternative programming paradigms such as OpenCL or Kokkos can also be employed.
This year not only classical High Performance Computing applications are targeted but special emphasis will be towards scientific applications that make use of artificial intelligence / deep learning methods (AI).
Various parallelization approaches are targeted for AI: data, tensor, and pipeline parallelism for any of the frameworks (pytorch, pytorch lighting, NeMo, etc.) The applications can do pre-training, post-training, finetuning, and / or inference.
EuroHack provides a unique opportunity for current or prospective users groups of large hybrid CPU-GPU systems to either (1) port their (potentially) scalable application to GPU accelerators, (2) optimize an existing GPU-enabled application, on a state-of-the-art GPU system, or (3) optimize for the multicore. Focus should be in any case the parallelism of the application. In case of AI applications a special emphasis will be the extension from single-GPU to multi-GPU / multi-node setups. The goal is that the development teams leave at the end of the week with applications executing faster, or at least with a clear roadmap of how to get there.
This program is addressed to small teams of 3 to maximum 4 developers interested in porting or optimizing their application on a cluster of CPUs and GPU accelerators in a short but extremely intense time window. This is a great opportunity for grad students and Postdocs.
Collectively the team should know the application intimately. There will be intensive mentoring during this 5-day hands-on workshop. Mentors come from universities and supercomputing centers such as CSCS and Swiss AI, as well as from industry such as NVIDIA, and they bring their extensive experience in programming GPGPUs, many of them contributing to develop the GPU-capable compilers and define the OpenACC standard. The in-person portion of the event will be limited to 5 or 6 teams of 3 to 4 developers with 2 mentors for each team.
CSCS is pleased to announce that submissions for the event EuroHack25 are now open. Team coordinators are invited to submit an application to be ported or optimized on GPUS or other massive parallel architectures by May 31, 2025 (AOE).
More information on the event page >