Applications: Make JUPITER shine
Applications are a central aspect in the exascale strategy of the Jülich Supercomputing Centre (JSC). Concrete use cases computed on JUPITER have the potential to advance developments in important areas of society and, in particular, to significantly accelerate change due to JUPITER's enormous total computing power. JUPITER will achieve more than 1 ExaFLOP/s for double-precision computing operations and more than 70 ExaFLOP/s for 8-bit precision, relevant for AI applications in particular.
JUPITER Benchmark Suite
Applications played an important role in the JUPITER procurement process. The JUPITER Benchmark Suite was developed to bring a system into production that not only performs very well in synthetic benchmarks but also in real and relevant use cases. Based on current and projected important applications, a collection of 16 applications was compiled and used for the performance evaluation of possible JUPITER configurations. Andreas Herten, lead author of the corresponding scientific paper, comments: “With the JUPITER Benchmark Suite, we have created a milestone for the a-priori evaluation of supercomputers. The combination of standardization and real applications is unique and allowed a targeted projection of the expected performance.”
JUPITER Research and Early Access Program
For applications, JUPITER is not only an opportunity but also a challenge. To use a supercomputer of such size to its full potential, applications must be thoroughly prepared and optimized. This can range from making adjustments to algorithms and methods to optimizing individual settings. Such optimizations are currently being carried out as part of the JUPITER Research and Early Access Program (JUREAP). In JUREAP, experts from JSC are working with more than 100 applications to assess and prepare them for exascale. The applications are highly diverse and have the potential to fundamentally advance many different scientific disciplines. They range from AI applications, such as foundation models and generative video generation, to climate models, particle physics and energy applications, to molecular dynamics simulations relevant for drug development and disease control. Some examples are briefly presented below.
Pioneering Exascale
Another important milestone on the journey towards exascale computing was the GCS Exascale Pioneer Call. The GCS Exascale Pioneer Call provided the first opportunity for applications to compete for computing time on JUPITER. To do so, applications had to demonstrate their exascale potential and present a convincing exascale use case. This also had an impact on the operation of the current JSC supercomputers. In order to give exascale candidates a fair chance to demonstrate their performance on a large number of GPUs, so-called Big Days were regularly offered at JSC. Big Days are a unique opportunity to use supercomputers at JSC at maximum scale. For the last Big Day in October, just before the submission deadline for the GCS Exascale Pioneer Call, demand was so high that it almost became a Big Week. Overall, JSC is very satisfied with the preparations of important exascale applications. Mathis Bode, head of JUREAP, summarizes: “Although evaluation of applications was very thorough and rigorous, the demand for computing time from eligible applications significantly exceeded the supply. We are very pleased with the applications and are excited to continue preparing them for JUPITER and ultimately to unleash JUPITER's full potential.”
JUPITER USE CASES
Biology - Biophysics
Prof. Dr. Gerhard Hummer, Director Theoretical Biophysics Department, Max Planck Institute of Biophysics
We use biomolecular simulations and modeling to investigate how biological systems operate at the molecular level, advancing our understanding of how living cells work as a basis for new therapies and nanotechnology applications. With JUPITER, we aim to perform molecular dynamics simulations of the nuclear pore complex - the largest protein assembly in cells - to reveal how it regulates molecular transport. With hundreds of millions of atoms, these simulations need exascale computing to reach meaningful timescales. Exascale offers unique atomistic insights, advancing nuclear transport models and paving the way for improved control of gene trafficking and the fight against retroviruses like HIV.
Medicine - Cellular Neuroscience
Prof. Dr. Katrin Amunts, Director of the Institute of Neuroscience and Medicine (INM-1) at Forschungszentrum Jülich
Our goal is to build a large foundational model of the human brain. Exascale will allow us to capture its complex structure down to the level of cells and their connections, precisely mapped to the anatomy of the brain. For the first time, JUPITER will allow us to link the treasure trove of image data collected over decades with results from other imaging or optical techniques with microscopic precision. This will allow us to gain new insights into the relationship between brain structure and function, and open up entirely new perspectives for investigating the relationship between neural network architecture and intelligence.
Physics - Nuclear and Elementary Particle Physics, Quantum Mechanics, Relativity, Fields
Prof. Dr. Zoltan Fodor, University of Wuppertal
As particle physicists we are computing microphysical quantities, like the magnetic moment of an elementary particle, called muon. JUPITER will be more than ten times larger than its predecessor, enabling us to substantially increase the resolution of our ``microscope'' and to reach unprecedented precision. The results of these computations can be compared to currently running experimental measurements. If they differ, it will have profound consequences on the fundamental laws of Nature, like the existence of a not yet discovered particle or a presence of a new interaction.
Prof. Dr. Christoph Lehner, University of Regensburg
We investigate the physics of quarks and gluons from very low to very high energies. With JUPITER, we for the first time have a machine that allows us to study spacetime boxes that are sufficiently large so that they can contain at the same time the long distance scales where quarks and gluons are bound but also shorter distance scales where they are quasi free. Such a setup enables new observables to be studied for which both aspects are important but is also crucial for high precision for some of our lower-energy research targets. This work is only possible with an exascale supercomputer and we are eager to fully utilise JUPITER to allow us to enter this exciting new chapter of our research.
Prof. Dr. Dr.h.c. Ulf-G. Meißner, Universität Bonn and Forschungszentrum Jülich
Understanding how nuclei are formed and how the heavy elements are generated is still one of the unsolved problems of fundamental physics. With JUPITER, we want to perform large scale simulations using nuclear lattice effective field theory to explore the limits of nuclear stability, to calculate nuclear reactions relevant to the element generation and explore the properties of the densest objects in the universe, namely neutron stars. This is only possible with exascale computing because of the large number of involved particles and the volume sizes to sufficiently suppress lattice artefacts. In addition, these methods will pave the way for improved material simulations and brain research beyond the mean-field approximation.
Prof. Dr. Kristel Michielsen, Head of the Research Group Quantum information Processing and the Jülich UNified Infrastructure for Quantum computing ― JUNIQ at the Jülich Supercomputing Centre: “We perform large-scale simulations of quantum computers and annealers and benchmark and study prototype applications for this new compute technology. With JUPITER we plan to emulate a 50-qubit gate-based universal quantum computer, and a compiled version of Shor's algorithm and quantum annealing of the Hubbard model on a gate-based universal quantum computer. As the GH200 superchips of JUPITER allow the CPU and GPU memory to be exposed to both the GPU and CPU of each chip, it is possible to effectively double the amount of available memory, allowing us to set the 50 qubit world record. Simulations of quantum computers of that size allow studying the performance of potential applications that will significantly influence society.
Physics - Statistical Physics, Soft Matter, Biological Physics, Nonlinear Dynamics
Prof. Gerhard Gompper, director of Institute IAS-2 —Theoretical Physics of Living Matter — at Forschungszentrum Jülich, and faculty member of the Institute for Biological Physics at University of Cologne
We study the behavior of biological fluids, like blood cells in the blood stream, as well as active microorganisms, e.g., bacteria, algae, and sperm, in fluid environments. In the latter systems, the directed and self-steering swimming motion of the organisms gives rise to a multitude of dynamic self-organizations, such as maritime algal blooms and bacterial turbulence, where structure formation occurs on length scales much larger than the microorganisms themselves. Therefore, the simulations require an enormous computing power, as it can only be provided by exascale computers like JUPITER. Our studies may contribute to the design of swarming microrobots, and to biotechnology for carbon sequestration and biofuels.
Physics - Astrophysics and Astronomy
Prof. Dr. Marcus Brüggen, Professor for Extragalactic Astrophysics, University of Hamburg
We are concerned with the fundamental properties of magnetized turbulence. With JUPITER, we want to perform the largest magnetohydrodynamics turbulence simulation to date to determine if an asymptotic regime exists, and, if so, what it looks like. This simulation is only possible with exascale resources due to the extreme dynamical range required, which translates to unprecedented needs in combined computing, memory, and storage. We are excited to see if the results match any existing theories or whether new approaches are required – supported by data and potentially contributing to fusion research to solve the energy crisis.
Geosciences - Atmospheric Science
Dr. Daniel Klocke, Max-Planck-Institut für Meteorologie
Climate change poses profound questions for society, creating fascinating puzzles for scientists. Central to these questions and puzzles is how the small connects to the large, and the large to the small. Do storms matter for the global pattern of winds, and given the global pattern of winds, what can we say about storms. In the past, the puzzles seemed insolvable, because the scales spanned were incomputable. JUPITER opens a new chapter in efforts to solve these puzzles, by simulating the global climate with local granularity. Solving the puzzles will help society answer its questions as to which changes in climate are inevitable, and how best to prepare for them.
Dr. Nils Peter Wedi, Digital Technology Lead of Destination Earth at the European Centre for Medium-Range Weather Forecasts (ECMWF)
Supercomputing is intimately connected with our ability to accurately simulate weather extremes in a warming climate and to project to observed regional and local scale changes in our environment. Together with researchers across many European and national institutions we implement ‘Destination Earth’, extreme scale digital twin simulations of our Earth system to do just that. We are excited to have Europe’s largest supercomputer JUPITER in the immediate vicinity of our new campus in Bonn. These exascale computing resources are required to provide us with quantitative uncertainty bounds and support decision making on a range of adaptation measures in areas such as agriculture, health, and energy.
Thermal Engineering/Process Engineering - Fluid Mechanics
Prof. Dr. Andrea Beck, Deputy Director of the Institute of Aerodynamics
Reducing the environmental footprint of air traffic requires radical new design ideas for airframes and propulsion, driven and enabled by a deeper understanding of the intricate flow physics. JUPITER will enable us to investigate the coupled problem of shock / boundary layer interaction on a wing in realistic flight conditions through extremely detailed simulations that resolve all relevant interacting phenomena - this is only possible on an exascale system due to vast rage of length and time scales involved. Through the simulations on JUPITER we will be able to understand and predict these flows: A key step in engineering cleaner, quieter and safer aircraft.
Prof. Dr.-Ing. Christian Hasse, Head of the department Simulation of reactive Thermo-Fluid Systems at Technical University of Darmstadt
Replacing fossil fuels with hydrogen, e.g., in CO₂-free power generation, is essential for achieving net zero emissions. However, the dynamics of turbulent, pressurized hydrogen flames differ significantly from those of conventional fuels and remain poorly understood. Exascale supercomputing is driving progress in fundamental science, it enables us to unravel the complexities of hydrogen combustion dynamics through direct numerical simulation (DNS). We can now resolve the shortest timescales down to nanoseconds and capture turbulent flame structures at the micrometer scale. With JUPITER, for the first time, exascale computing enables DNS under technically relevant gas turbine conditions—a milestone in Computational Engineering and Process Virtualization.
Prof. Dr. Jörg Schumacher, Head of the Fluid Mechanics Group at TU Ilmenau
Buoyancy-driven turbulent convection flows are ubiquitous in nature and technology ranging from stellar interiors to blanket cooling in nuclear fusion reactors. Rayleigh-Bénard convection, a plane fluid layer between two plates uniformly heated from below and cooled from above, is the paradigm for all these multi-physics processes. The new JUPITER supercomputer will give us the opportunity to gain a deeper understanding of the dynamics of the coupled thermal and viscous boundary layers which determine the global heat transfer. The coming generation of exascale computations will reveal unprecedented details on the structures in natural convection and their connection to statistical fluctuations of the fields at turbulence levels never obtained before.
Computer Science, Systems and Electrical Engineering - Interactive and Intelligent Systems, Image and Language Processing, Computer Graphics and Visualisation
Prof. Dr. Björn Ommer, Head of Computer Vision and Learning Group, LMU Munich
Our research focuses on lightweight foundation models for video representation and generation, inspired by the success of generative approaches like Stable Diffusion. Through JUPITER, we aim to develop spatio-temporal compression and diffusion architectures that enable the creation of high-quality, accessible video models. Exascale computing plays a pivotal role, allowing us to train on vast datasets while optimizing model efficiency. This approach opens exciting possibilities for developing video models that generalize far beyond their training data. Such models have the potential to drive societal impact across diverse domains—from advancing medical imaging to improving autonomous driving—while fostering accessibility and innovation.
Contact
- Institute for Advanced Simulation (IAS)
- Jülich Supercomputing Centre (JSC)
Room 307
Dr. Andreas Herten
Co-Lead of division Novel System Architecture design, head of ATML Accelerating Devices
- Institute for Advanced Simulation (IAS)
- Jülich Supercomputing Centre (JSC)
Room 228