Europe’s First Exascale Supercomputer: JUPITER’s Groundbreaking Year
JUPITER, Europe’s inaugural exascale supercomputer located at Germany’s Forschungszentrum Jülich, has made significant strides in advanced computing over the past year. Powered by NVIDIA Grace Hopper Superchips and NVIDIA Quantum-X800 InfiniBand networking, JUPITER is at the forefront of four pioneering projects showcased during the International Supercomputing Conference (ISC) in Hamburg. These initiatives aim to tackle complex scientific challenges, including mapping the human brain, simulating Earth’s climate, developing AI for next-gen wireless networks, and modeling a universal quantum computer.
Pioneering Brain Mapping with CytoNet
The Jülich Brain Atlas project is a collaborative effort involving the Institute of Neuroscience and Medicine at Jülich, Helmholtz AI, and various partner institutions. The project has yielded CytoNet, a foundation model designed for analyzing brain microarchitecture. The human brain’s complexity—comprising 86 billion neurons interconnected by approximately 100 trillion synapses—has historically posed challenges for researchers aiming to understand its functions at a granular level.
Led by neuroscientist Katrin Amunts and computer scientist Christian Schiffer, the research utilizes brain imaging data to create a comprehensive map linking individual cell structures to broader organizational patterns in the brain. Training this model on JUPITER took less than five days, processing 6.5 petabytes of data derived from 21 post-mortem brains using 4,096 NVIDIA Grace Hopper Superchips. A detailed paper on this groundbreaking work is available on arXiv.
Amunts highlighted the significance of this research, stating that it marks a paradigm shift in neuroscience by not only using AI for analysis but also creating an intelligent agent capable of conducting experiments autonomously. Future developments will focus on integrating multimodal reasoning and language interfaces into AI agents that assist researchers in interrogating brain data directly.
Climate Simulation at Unprecedented Resolution
A collaborative team from ETH Zurich, German Climate Computing Centre (DKRZ), Jülich Supercomputing Centre (JSC), Max Planck Institute for Meteorology, NVIDIA, Swiss National Supercomputing Centre (CSCS), and the University of Hamburg has developed an innovative configuration of the ICON climate model. This model won the prestigious Gordon Bell Prize for Climate Modelling at SC25 last November.
What sets ICON apart is its capability to simulate a coupled Earth system at an unprecedented 1-kilometer resolution across oceanic, atmospheric, land biogeochemistry, and carbon cycle components. This holistic approach allows for detailed simulations of ecosystems like phytoplankton blooms and zooplankton grazing—previous models could only address isolated aspects of these systems.
Utilizing 20,480 NVIDIA Grace Hopper Superchips on JUPITER, ICON achieved a remarkable feat: simulating approximately 146 days of real climate within just 24 hours of computation time. Daniel Klocke from the Max Planck Institute emphasized that this level of detail enables scientists to observe interactions among atmospheric conditions, ocean dynamics, and biological processes directly resulting from physical laws rather than approximations.
Advancing Wireless Networks through AI
This March marked a significant collaboration between Ericsson and Forschungszentrum Jülich aimed at enhancing AI capabilities for next-generation wireless networks (5G and beyond). With JUPITER serving as the computational backbone for large-scale AI model training and testing, this partnership seeks to develop brain-inspired architectures that can manage complex network operations with greater energy efficiency.
The research will focus on creating AI models tailored for Ericsson’s radio and core networks while exploring energy-efficient AI inference methods at the radio edge through neuromorphic approaches. Additionally, concepts derived from JSC’s exascale work will inform modular supercomputing architectures that enhance network performance.
Setting New Records in Quantum Computing
Researchers at the Jülich Supercomputing Centre have achieved a landmark breakthrough by fully simulating a universal 50-qubit quantum computer—surpassing the previous record of 48 qubits. This accomplishment was made possible by leveraging the tightly integrated CPU-GPU memory architecture inherent in JUPITER’s NVIDIA GH200 Grace Hopper Superchips. This architecture allows data exceeding GPU memory limits to spill into CPU memory seamlessly without significant performance loss.
The JUQCS-50 quantum simulator serves as an essential tool for quantum research as current quantum hardware struggles to outperform classical computers on practical problems. By simulating larger quantum states than previously possible, researchers can design and stress-test algorithms intended for future quantum machines. JUQCS-50 will be accessible through JUNIQ—the quantum computer user facility led by Kristel Michielsen at JSC—providing valuable insights into algorithm performance as Europe advances toward more sophisticated quantum-GPU supercomputers.
The Broader Implications of Exascale Computing
The diverse range of scientific projects underway on JUPITER—from neuroscience to climate modeling to advancements in wireless technology and quantum computing—demonstrates that exascale computing has transitioned from theoretical research into practical applications with tangible benefits. The results achieved thus far validate the capabilities of NVIDIA’s Grace Hopper platform as it continues to push boundaries across various scientific disciplines.
What This Means
The advancements facilitated by JUPITER underscore Europe’s leadership in exascale computing while addressing some of humanity’s most pressing scientific challenges. As these projects progress, they promise not only to enhance understanding in their respective fields but also to pave the way for future innovations that could significantly impact technology and society at large.
For more information, read the original report here.
