Title

Post-Fugaku Projection
-Where will industrial applications of HPC/AI go in the post-Fugaku era?-

Abstract

Following the successful completion of its feasibility study, the Fugaku-next project has recently been initiated. With its completion expected in approximately five years, the "post-Fugaku" era is fast approaching. Drawing on over forty years of experience in the development and industrial application of HPC software, this talk provides a strategic perspective on the future of the field.

The presentation begins by summarizing the milestone industrial applications achieved for the first time by Fugaku. It then introduces the architecture of Fugaku-next, specifically highlighting its compute nodes. Empowered by unprecedented GPU capabilities, these nodes are expected to deliver a 100-fold increase in FP64 performance per node.

This leap in capability suggests that benchmark computations currently requiring 10,000 Fugaku nodes could be performed using only 100 nodes in the post-Fugaku era—provided that application software achieves a corresponding 100-fold increase in sustained performance. Consequently, the transition from large-scale benchmark computations to routine production runs is poised to be a defining characteristic of this new era. The talk concludes by identifying the challenges of application speedup, advocating for a robust co-design approach to boost collaboration among hardware designers, system-software architects, and application developers.

Bio

After graduating from the Mechanical Engineering Department of the University of Tokyo, Dr. Chisachi Kato began his career in 1984 at the Mechanical Engineering Research Laboratory of Hitachi, Ltd., where he served as a research engineer for 15 years. In January 1999, he joined the Institute of Industrial Science (IIS) at the University of Tokyo, where he was appointed Professor in 2003. During his tenure, he served as the Director of the Center for Research on Innovative Simulation Software (CISS) until March 2023. Following his retirement from the University of Tokyo, where he remains Professor Emeritus, he joined Nihon University in April 2024, currently serving as a Research Project Professor.

His research focuses on large-scale numerical simulation of fluid flows and aeroacoustics, characterized by extensive collaboration with industrial partners. Dr. Kato has led seven national projects dedicated to developing and distributing cutting-edge simulation software optimized for high-performance computing (HPC). This software, released as freeware through CISS, has reached a significant milestone of over 100,000 downloads, bridging the gap between academic innovation and industrial application.

BACK TO TOP

Title

Non-Orthogonal Configuration Interaction with Fragments at Exascale

Abstract

Exascale supercomputers enable implementation and application of advanced computational chemistry methodologies for molecular systems of realistic size and higher accuracy. Among these is the non-orthogonal configuration interaction with fragment wave functions (NOCI-F) approach which has been implemented in the massively parallel and accelerated open-source software GronOR. Wave functions are expressed as expansions in terms of many-electron basis functions (MEBFs) constructed as spin-adapted anti-symmetrized products of independently optimized multi-configuration fragment wave functions. Advantage is the intuitive description of molecular spin states by relatively short expansions in terms of individual fragment wave functions, important for the description of many molecular energy and exciton transfer processes.

Orbital sets of the MEBFs are mutually non-orthogonal, resulting in compute-intensive Hamiltonian matrix element evaluations. Dynamic correlation is included through shifting of diagonal matrix elements of the NOCI matrix, or by explicit incorporation in the fragment wave functions. Computational complexity can be reduced by transformation to common molecular orbitals. The implementation for exascale architectures is based on four design principles. First, the computational problem is mapped onto a task-based master-worker programming model minimizing data communication and load imbalance. Second, the code is using directive-based accelerator off-loading to GPU devices using OpenACC or OpenMP, maintaining portability between architectures. Third, the programming approach provides resiliency to certain classes of hardware faults. Fourth, the software implementation supports execution on heterogeneous computer architectures.

Advantages are illustrated of application of NOCI-F calculations to the process of singlet fission in photovoltaic materials, and capabilities and implementation details highlighted, with particular attention to parallel and accelerated efficiencies on computational resources up to the largest supercomputers such as Frontier at the Oak Ridge Leadership Computing Facility.

Acknowledgment
This work used resources through the INCITE program of the Oak Ridge Leadership Computing Facility (OLCF) at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.

Reference
T. P. Straatsma, R. Broer, A. Sánchez-Mansilla, C. Sousa, C. de Graaf, GronOR: Scalable and Accelerated Nonorthogonal Configuration Interaction for Molecular Fragment Wave Functions, J. Chem. Theory Comp. 18, 3549-3565 (2022).

Bio

Dr. Tjerk P. Straatsma is an internationally recognized scientist with more than 45 years of experience in the development, efficient implementation and application of advanced modeling and simulation methods as key scientific tools in the study of chemical and biomolecular systems, complementing analytical theories and experimental studies. His research focuses on the development of computational techniques that provide unique and detailed atomic level information that is difficult or impossible to obtain by other methods, and that contributes to the understanding of the properties and function of these systems.

His research interests are in the combination of the design of efficient implementations of these methods on modern, complex computer architectures, including the vector processing supercomputers of the 1980s and the massively parallel computer systems of today. He has applied biomolecular simulation methods to study a variety of systems of interest, such as microbial membrane mediated geochemical processes at mineral surfaces, material and signal transport processes across microbial membranes, and complex enzymatic reactions that involve combined electron and proton transfer steps.

Dr. Straatsma earned his Ph.D. in Mathematics and Natural Sciences from the University of Groningen, and moved to the University of Houston in 1988 as a Postdoctoral Research Associate. In 1990 he joined the Faculty as Research Assistant Professor in the Department of Chemistry.

Dr. Straatsma joined Pacific Northwest National Laboratory in 1995, where he established the Computational Biology and Bioinformatics research group, and advanced to the rank of Laboratory Fellow. He served as Director for the eXtreme Scale Computing Initiative (XSCI) to build the capabilities needed to enable scientific advancements and breakthroughs in selected domain sciences through computational modeling and simulation on next-generation, extreme-scale computers.

Dr. Straatsma joined Oak Ridge National Laboratory in 2013 as manager of the Scientific Computing group in the National Center for Computational Sciences. This center is the site for the Oak Ridge Leadership Computing Facility and houses the largest supercomputer for open science in the United States. He currently is Chief Scientist in the Leadership Team of the National Center for Computational Sciences.

Dr. Straatsma is a core developer for the NWChem and NWChemEx massively parallel computational chemistry software suite, and the lead developer for the ARGOS molecular dynamics simulation package, and the GronOR massively parallel and GPU-accelerated Non-Orthogonal Configuration Interaction code.

BACK TO TOP

Title

Designing High-Performance, Scalable, and Converged Middleware for the Modern HPC and AI Systems

Abstract

This talk focuses on challenges and opportunities in designing middleware for HPC and AI (Deep/Machine Learning) workloads on modern high-end computing systems. The talk initially presents the challenges in co-designing HPC software by considering support for dense multi-core CPUs, high-performance interconnects, GPUs, and DPUs. Advanced designs and solutions (such as RDMA, in-network computing, GPUDirect RDMA, on-the-fly compression) to exploit novel features of these emerging technologies and their benefits in the context of MVAPICH libraries (http://mvapich.cse.ohio-state.edu) are presented. Next, the talk focuses on MPI-driven converged software stack for the AI (Deep/Machine Learning) domains to extract performance and scalability for popular Deep Learning frameworks, large out-of-core models, and GPUs (http://hidl.cse.ohio-state.edu).

Bio

DK Panda is a Professor and University Distinguished Scholar of Computer Science and Engineering at the Ohio State University. He is serving as the Director of the ICICLE NSF-AI Institute (https://icicle.ai). He has published over 500 papers. The MVAPICH MPI libraries, designed and developed by his research group (http://mvapich.cse.ohio-state.edu), are currently being used by more than 3,475 organizations worldwide (in 93 countries). More than 1.99 million downloads of this software have taken place from the project's site. This software is empowering many clusters in the TOP500 list. High-performance and scalable solutions for Deep Learning frameworks and Machine Learning applications from his group are available from https://hidl.cse.ohio-state.edu. Similarly, scalable, and high-performance solutions for Big Data and Data science frameworks are available from https://hibd.cse.ohio-state.edu. Prof. Panda is a Fellow of ACM and IEEE. He is a recipient of the 2022 IEEE Charles Babbage Award and the 2024 IEEE TCPP Outstanding Service and Contributions Award. More details about Prof. Panda are available at http://www.cse.ohio-state.edu/~panda.

BACK TO TOP

Title

Embodied AI for Science

BACK TO TOP

Title

AI-Assisted Code Generation for HPC: Lessons from the HPC-GENIE Project

Abstract

The rapid diversification of high-performance computing (HPC) architectures has made performance-oriented software development increasingly complex. AI-assisted code generation has emerged as a promising approach to mitigating this burden; however, its practical effectiveness in real HPC workflows remains an open question. This invited talk presents lessons learned from the HPC-GENIE project, which investigates the use of generative AI for supporting HPC application development and performance tuning.

We introduce VibeCodeHPC, a prototype multi-AI-agent system that assists programmers through collaborative code generation, optimization suggestions, performance analysis, and iterative refinement. VibeCodeHPC integrates multiple specialized AI agents that emulate roles such as HPC programmer and code evaluator, enabling a human-in-the-loop code tuning process tailored to HPC environments.

The talk highlights concrete tuning scenarios from scientific computing, including matrix–matrix multiplication kernels, where AI agents guide loop transformations, parallelization strategies, and memory-access optimizations across MPI, OpenMP, and accelerator-based programming models. Through these case studies, we demonstrate how AI-assisted workflows can accelerate prototyping and expose performance bottlenecks, while still requiring expert knowledge for architecture-aware tuning and rigorous validation.

Based on these experiences, we discuss current limitations of AI assistance, such as limited hardware awareness and challenges in verification. We also present practical guidelines and outline future research directions toward trustworthy, performance-aware AI systems for next-generation HPC software development, including systems such as “FugakuNEXT.”

Bio

Takahiro Katagiri is a Professor at the Information Technology Center, Nagoya University. He received his B.S. degree from Kyoto University and his M.S. and Ph.D. degrees from the University of Tokyo. His academic career includes positions at the Japan Science and Technology Agency, the University of Electro-Communications, and the University of Tokyo. He has authored ten books and over 100 research articles. His research interests include high performance computing (HPC), software autotuning, hybrid HPC-quantum computing, and code generative AI. He has received several awards, including the Young Scientists’ Prize from Ministry of Education, Culture, Sports, Science and Technology, Japan.

BACK TO TOP

Title

R-CCS Breakthrough Future Computing in HPC/AI/Quantum

Abstract

2025 has been a landmark year for R-CCS in that we have firmly established your position as one of the global leaders in High Performance computing, driven and enabled by ModSim HPC, AI four Science, and Quantum-HPC hybrid computing. We will cover the current progress of the respective major national or semi-national projects that will enable breakthrough computing in HPC/AI/Quantum.

Bio

Satoshi Matsuoka has been the director of RIKEN Center for Computational Science (R-CCS) since 2018. He is responsible for developing the supercomputer Fugaku which has become the fastest supercomputer in the world in all four major supercomputer rankings in 2020 and 2021 (Top500, HPCG, HPL-AI, Graph500), along with multitudes of ongoing cutting edge HPC research being conducted, including investigating Post-Moore era computing, especially the future FugakuNEXT supercomputer.

He holds Ph. D. from the University of Tokyo in 1993. He was a Professor at the Global Scientific Information and Computing Center (GSIC), the Institute of Science Tokyo (formerly known as the Tokyo Institute of Technology), the director of the joint AIST-Tokyo Tech. Real World Big Data Computing Open Innovation Laboratory (RWBC-OIL). He was the leader of the TSUBAME series of supercomputers that had also received many international acclaims at the Institute of Science Tokyo, where he still holds a professor position, to continue his research activities in HPC as well as scalable Big Data and AI.

He has written over 500 articles according to Google Scholar, and chaired numerous conferences, including the Technical Papers Chair and the Program Chair for ACM/IEEE Supercomputing 2009 and 2013 (SC09 and SC13) respectively as well as many other conference chairs, and the ACM Gordon Bell Prize selection committee chair in 2018.

He is a Fellow of ACM, ISC, JSSST (Japan Society for Software Science and Technology) and IPSJ (Information Processing Society of Japan). His accolades are the ACM Gordon Bell Prizes in 2011 & 2021; the IEEE-CS Sidney Fernbach Award in 2014 as well as the IEEE-CS Computer Society Seymour Cray Computer Engineering Award in 2022, both being the highest awards in the field of HPC, and the only individual to receive both awards. He was selected as one of the HPCwire 35 Legends by the HPC wire in 2024.
His longtime contribution for the computer science research was commended with the Medal of Honor with Purple ribbon by His Majesty the Emperor of Japan in 2022, and he received the Public Service Medal (Friends of Singapore) at the Singapore National Day Awards in 2025.

BACK TO TOP

Title

Overview and Current Status of the FugakuNEXT Project

Abstract

RIKEN has been driving the development of the “FugakuNEXT” project since the beginning of 2025 and is now collaborating with Fujitsu Limited and NVIDIA to advance the basic design phase. For the first time in Japan’s flagship supercomputing systems, GPUs will be adopted as accelerators, with NVIDIA leading the design of the GPU infrastructure and Fujitsu leading the design of the overall system, compute nodes, and CPUs. The goal of the project is to achieve world-class performance in both simulation and AI, thereby delivering new value as a next-generation AI–HPC convergence platform. In this talk, we will present an overview and the current status of the project.

Bio

Kentaro Sano is the principal of the processor research team, and the leaders of the advanced AI device development unit and the next-generation high-performance computing infrastructure system development unit at RIKEN Center for Computational Science (R-CCS), responsible for research and development of future processors and systems for HPC and AI. He is also a visiting professor with an advanced computing system laboratory at Tohoku University. He received his Ph.D. from the graduate school of information sciences, Tohoku University, in 2000. Since 2000 until 2018, he was a Research Associate and an Associate Professor at Tohoku University. He was a visiting researcher at the Department of Computing, Imperial College, London, and Maxeler Technology corporation in 2006 and 2007. His research interests include data-driven and spatial-parallel processor architectures such as a coarse-grain reconfigurable array (CGRA) for HPC and AI, FPGA-based high-performance reconfigurable computing, quantum error correction hardware for fault-tolerant quantum computers, and system architectures for next-generation supercomputers including FugakuNEXT.

BACK TO TOP

Title

FugakuNEXT and FUJITSU-MONAKA Series

Abstract

The computing landscape is transitioning to a new era, one that merges numerical computation and AI for advanced simulation. This presentation outlines a vision for evolving High-Performance Computing (HPC) in the age of FugakuNEXT. We also introduce the Arm-based CPU, FUJITSU-MONAKA-X, intended as the processor for FugakuNEXT, alongside its predecessor, FUJITSU-MONAKA.

Bio

Yoshihiro Kusano is a Director at Advanced Technology Development Unit of Fujitsu, responsible for system development. Since joining Fujitsu, he has gained experience in the development of supercomputer and high-end servers, leading the development of the K-computer and Fugaku.

BACK TO TOP

Title

“Systems for Scientific Discovery in the Age of AI”

Abstract

Scientific computing is going through a transformation towards integrating dynamic, hybrid, and interactive elements with traditional ModSim. New AI methods, digital twins, and increased integration with external instruments can unlock leaps in discovery but achieving them will require thinking very differently about the architecture of systems, the way users interact with them, and the ways we measure success.

Bio

Dr. Daniel Ernst is the Senior Director of Supercomputing Technology at Nvidia, where he works on the intersection of accelerated computing architecture and high-performance computing. Prior to Nvidia, Dan has led technology pathfinding in multiple domains across HPC and cloud systems architecture, including leading the memory architecture team at Microsoft Azure and as principal investigator and architect for Cray’s exascale computing programs. Dan received his Ph.D. in Computer Science and Engineering from the University of Michigan, where he studied high-performance, low-power, and fault-tolerant microarchitectures.

BACK TO TOP