JUPITER: Europe’s First Exascale Supercomputer Showcases Groundbreaking Research
JUPITER, Europe’s inaugural exascale supercomputer located at Germany’s Forschungszentrum Jülich, has made significant strides in various scientific fields over the past year. Powered by NVIDIA Grace Hopper Superchips and NVIDIA Quantum-X800 InfiniBand networking, JUPITER is currently being highlighted at the International Supercomputing Conference (ISC) in Hamburg, where it supports four pioneering projects that illustrate the potential of exascale computing.
A Foundation Model for Mapping the Brain
The Jülich Brain Atlas project, in collaboration with Helmholtz AI and various partner institutions, has introduced CytoNet, a foundation model designed for analyzing brain microarchitecture. This initiative aims to unravel the complexities of the human brain, which contains approximately 86 billion neurons interconnected by around 100 trillion synapses.
Led by neuroscientist Katrin Amunts and computer scientist Christian Schiffer at Jülich’s Institute of Neuroscience and Medicine, this research leverages brain imaging data at a cellular scale. The model constructs a comprehensive map that correlates individual cell structures with broader patterns of brain organization and functionality.
Training for CytoNet was completed in under five days using 6.5 petabytes of data from 21 post-mortem brains across 4,096 NVIDIA Grace Hopper Superchips. A detailed paper on this groundbreaking work is available on arXiv. Amunts emphasized the transformative potential of this research: “For the first time, we’re not just using AI to analyze the brain — we’re building an agent that can think through the experiment itself.” The next phase involves developing an AI agent capable of assisting researchers in interrogating brain data directly.
Climate at Kilometer Resolution
A novel configuration of the ICON climate model has garnered attention for its ability to simulate a coupled Earth system at an unprecedented 1-kilometer resolution. Developed by a consortium including ETH Zurich and the Max Planck Institute for Meteorology, this model won the prestigious Gordon Bell Prize for Climate Modelling last November.
Unlike previous models that could only simulate isolated components of Earth’s systems, ICON integrates ocean, atmosphere, land, biogeochemistry, and carbon cycles into one cohesive framework. This comprehensive approach enables simulations that visualize entire ecosystems—such as phytoplankton blooms and zooplankton grazing—at a level of detail previously unattainable.
Running on 20,480 NVIDIA Grace Hopper Superchips on JUPITER, ICON achieved a world record by simulating roughly 146 days of real climate within just 24 hours of computation. Daniel Klocke from the Max Planck Institute noted that this high-resolution simulation allows scientists to observe interactions among atmospheric, oceanic, and biospheric processes directly from fundamental physical laws.
6G Gets an Exascale Partner
In March 2023, Ericsson partnered with Forschungszentrum Jülich to leverage JUPITER as a computational engine for advancing AI technologies aimed at enhancing both 5G and future 6G networks. This collaboration focuses on developing brain-inspired architectures intended to optimize complex network operations while minimizing energy consumption.
The research agenda encompasses AI models tailored for Ericsson’s radio and core networks as well as energy-efficient AI inference strategies utilizing neuromorphic approaches. By drawing from insights gained through exascale computing efforts at Jülich, this initiative aims to innovate modular supercomputing architectures that can better support next-generation telecommunications infrastructure.
Breaking Quantum Records
Researchers at the Jülich Supercomputing Centre have set a new benchmark by fully simulating a universal 50-qubit quantum computer. This achievement surpasses the previous record of 48 qubits and showcases JUPITER’s capabilities in quantum research.
The successful simulation was made possible through JUPITER’s advanced CPU-GPU memory architecture which allows data exceeding GPU limits to seamlessly transition into CPU memory without significant performance loss. This capability enabled researchers to maintain a greater quantum state than what was previously achievable with GPU memory alone.
This powerful quantum simulator—dubbed JUQCS-50—will serve as a vital resource for exploring quantum algorithm designs within JUNIQ, the quantum computing user facility at JSC. Kristel Michielsen, director of JSC, highlighted JUQCS-50’s role as a testbed for future quantum-GPU supercomputers.
Exascale’s Impact
The diverse range of scientific endeavors being conducted on JUPITER—from neuroscience to climate modeling to telecommunications and quantum computing—underscores how exascale computing has transitioned from theoretical concepts into practical applications. The results achieved thus far serve as compelling evidence for the effectiveness of NVIDIA’s Grace Hopper platform in pushing scientific boundaries.
What This Means
The advancements enabled by JUPITER illustrate not only Europe’s leadership in exascale computing but also its potential impact across multiple scientific disciplines. As researchers continue to harness this immense computational power, breakthroughs in understanding complex systems—from human cognition to climate dynamics—are anticipated to accelerate significantly. The implications extend beyond academia; industries reliant on cutting-edge technology will likely benefit from these innovations in ways yet to be fully realized.
For more information, read the original report here.
