The International Conference for High Performance Computing, Networking, Storage, and Analysis
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Authors: Amit Ruhela (Texas Advanced Computing Center (TACC), University of Texas), David Martin (Northwestern University), Steffen Christgau (Zuse Institute Berlin), Hatem Ltaief (King Abdullah University of Science and Technology (KAUST)), Clayton Hughes (Sandia National Laboratories), Nalini Kumar (Intel Corporation), David Keyes (King Abdullah University of Science & Technology), Andreas Herten (Forschungszentrum Jülich), Antonio Peña (Barcelona Supercomputing Center (BSC)), William Fowler (Intel Corporation)
Abstract: With the increasing demand for AI in HPC, there has been a rapid rise in accelerated architectures, portable programming models, and frameworks. The already-daunting task of programming for accelerated systems has become even more complex. This BoF, organized by IXPUG, will focus on portable programming across a wide range of heterogeneous architectures—including Intel, NVIDIA, AMD, and Arm—supporting diverse simulation, data analytics, and AI workloads. The session will explore key challenges, state-of-the-art solutions, and emerging best practices for programming across these systems, identifying common principles and methodologies that support development and long-term maintenance across sites, architectures, and scientific applications.
Long Description: IXPUG is a global community of hundreds of application users, software developers, and HPC center staff who leverage HPC hardware and software technologies—including processors, accelerators, memory, software tools, and storage middleware—to address some of the world’s most complex computational challenges on some of the world's most powerful supercomputers. This BoF, organized by IXPUG but vendor-neutral, will focus on sharing experiences and expertise in programming across a wide spectrum of accelerated architectures (including AI accelerators), targeting a diverse set of workloads spanning simulation, data analytics, AI, and composite applications.
As AI techniques are increasingly integrated into traditional HPC codes, the community anticipates even greater system diversity in the years ahead. This trend underscores the urgent need for programming models and abstractions that support efficient development across a variety of architectures and application domains. Application developers and computational scientists face growing challenges due to differences in hardware designs, parallelization strategies, hardware primitives, and architectural features, all of which complicate the task of creating portable and efficient software capable of running across diverse machines and at multiple HPC sites.
This BoF will delve into current approaches, state-of-the-art techniques, and emerging best practices for programming heterogeneous and novel architectures. The objective is to identify common principles and methodologies that enable sustainable and portable development across systems, sites, and scientific applications. The session will highlight real-world experiences from developers using modern programming frameworks such as Kokkos, RAJA, PyTourch, and oneAPI, as well as from application users who have benchmarked performance across a wide variety of systems.
Through invited talks by researchers from HPC centers and developers of widely-used programming abstractions, this BoF aims to provide a rich forum for exchanging practical insights and technical expertise. Participants—including researchers, application developers, HPC facility staff, and industry stakeholders—will be encouraged to engage with the speakers and each other to explore effective strategies for dealing with heterogeneous computing environments and the increasingly varied nature of today’s workloads.
The format will consist of concise presentations during the first half of the session, followed by a moderated discussion to foster interactive dialogue among speakers and attendees. This structure will encourage a balanced exchange of ideas, real-world experiences, and anticipated challenges, as well as potential solutions and future directions.
This BoF continues a long tradition of successful IXPUG-hosted sessions, starting at SC14 and continuing through SC23, consistently attracting 80–150 attendees annually. IXPUG has also organized related BoFs and workshops at ISC High Performance and HPC Asia. For this session, the IXPUG steering committee has assembled a diverse group of experts who will present their insights on portable programming models in use today, along with the challenges and opportunities anticipated in the evolving HPC-AI landscape. A concluding discussion, led by IXPUG steering group officers, will allow attendees to engage directly with the speakers and share their own experiences and ideas.
Website: https://www.ixpug.org/events/sc25-bof