Europe’s first exascale supercomputer JUPITER, built on NVIDIA Grace Hopper Superchips and hosted at Germany’s Forschungszentrum Jülich, demonstrated four exascale science workloads at the 2025 ISC High Performance conference blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. The system’s 1 exaflop peak compute capacity confirmed that billion-billion-per-second exascale compute is already delivering on long-promised research breakthroughs across neuroscience, climate science, telecom, and quantum research, per NVIDIA’s official JUPITER science announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
JUPITER, Europe’s first exascale supercomputer, showcased its completed science workloads at ISC 2025 in Hamburg, Germany. Credit: NVIDIA
JUPITER Exascale Supercomputer Debuts Four High-Impact Science Workloads at ISC 2025
The four highlighted projects target scientific problems that were computationally intractable on pre-exascale supercomputers, per NVIDIA’s official JUPITER science announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. Three of the four projects have already run to completion on JUPITER, delivering validated, reproducible results for the global research community, while the fourth is an announced future collaboration that will leverage the system’s 1-exaflop peak capacity for upcoming 6G telecom and quantum research workloads. All four projects demonstrate that exascale compute is no longer a theoretical milestone, but a production tool delivering measurable scientific progress across neuroscience, climate science, telecom, and quantum physics.
Cellular-Scale Brain Mapping Cuts Training Time to Under Five Days
The Jülich Brain Atlas project, led by neuroscientist Katrin Amunts and computer scientist Christian Schiffer at the Institute of Neuroscience and Medicine (INM-1), released CytoNet, a foundation model for brain microarchitecture analysis blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. The model learns from cellular-scale brain imaging data to map individual neuron structures to broader patterns of brain organization and function.
Training ran on 4,096 NVIDIA Grace Hopper Superchips across JUPITER, processing 6.5 petabytes of data from 21 post-mortem brains in under five days blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. That timeline is a fraction of the months required to process the same dataset on pre-exascale supercomputers, per NVIDIA’s announcement.
The team’s next step is an AI research agent built on open models including NVIDIA Nemotron 3 120B, designed to let scientists interrogate brain data via natural language queries blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. For example, researchers can use the agent to cross-reference neuron structure patterns associated with motor function across all 21 brains without manual data filtering, per NVIDIA’s official announcement.
1-Kilometer Climate Simulation Sets Global Compute Record
A collaborative ICON climate model configuration, developed by researchers at ETH Zurich, the German Climate Computing Centre (DKRZ), Jülich Supercomputing Centre, Max Planck Institute for Meteorology, NVIDIA, Swiss National Supercomputing Centre (CSCS) and the University of Hamburg, won the 2025 Gordon Bell Prize for Climate Modelling for its work on JUPITER blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. It is the first model to simulate a fully coupled Earth system — including atmosphere, ocean, land, biogeochemistry and a full carbon cycle with cross-component carbon exchange — at 1-kilometer resolution, per NVIDIA’s official announcement.
This resolution enables direct simulation of fine-scale ecosystem processes such as phytoplankton blooms that were previously approximated via parameterized estimates in lower-resolution models. Running on 20,480 NVIDIA Grace Hopper Superchips, the model simulated 146 days of real-world climate in 24 hours of wall-clock compute time, a world record for global climate simulation blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
Daniel Klocke, computational infrastructure and model development group leader at the Max Planck Institute for Meteorology, noted the 1-kilometer resolution lets fine-scale winds, ocean eddies and upper-ocean mixing emerge directly from physical laws rather than parameterized approximations, giving unprecedented insight into carbon cycle regulation, per NVIDIA’s announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
6G AI and 50-Qubit Quantum Simulation Push Hardware Boundaries
In March 2025, Ericsson and Forschungszentrum Jülich announced a collaboration to use JUPITER as a future compute engine for large-scale AI model training and testing for 5G network evolution and 6G development, targeting brain-inspired, low-energy network architectures blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. The collaboration has not yet reported completed runs on the system, per NVIDIA’s official announcement.
Research priorities for the partnership include AI models for Ericsson’s radio and core networks, energy-efficient neuromorphic inference at the radio edge, and modular supercomputing architecture concepts drawn from JSC’s exascale deployment experience blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
Separately, JSC researchers working with the jointly run NVIDIA Application Lab set a world first by fully simulating a universal 50-qubit quantum computer, surpassing the prior 48-qubit record by 2 qubits blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/. The simulation leveraged the coherent, tightly coupled CPU-GPU memory architecture of JUPITER’s GH200 Grace Hopper Superchips, which lets quantum state data exceeding GPU memory limits spill seamlessly into CPU memory with minimal performance loss. This capability made the 50-qubit simulation possible where prior GPU-only systems hit a hard memory ceiling at 48 qubits, per NVIDIA’s announcement.
The team notes these classical simulations remain the most powerful tool for quantum algorithm research, as current quantum hardware cannot yet outperform classical computers on useful, real-world problems, per NVIDIA’s official JUPITER science showcase blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
Frequently Asked Questions About the JUPITER Exascale Supercomputer
What is the peak compute capacity of the JUPITER exascale supercomputer?
JUPITER delivers a peak compute capacity of 1 exaflop, equal to 1 billion billion calculations per second, per NVIDIA’s official ISC 2025 science announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
Which JUPITER exascale science workloads have already run to completion?
Three of the four demonstrated workloads have finished runs: the Jülich Brain Atlas CytoNet foundation model, the 1-kilometer resolution ICON climate simulation, and the 50-qubit quantum computer simulation. The fourth, the Ericsson 6G AI collaboration announced in March 2025, is an upcoming project with no completed runs yet, per NVIDIA’s announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
What resolution did the award-winning JUPITER climate simulation achieve?
The ICON climate model simulated a fully coupled Earth system at 1-kilometer resolution, the first global climate simulation to hit that granularity, per NVIDIA’s official announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
How many qubits did JUPITER’s quantum simulation surpass?
JUPITER’s full simulation of a universal 50-qubit quantum computer beat the prior world record of 48 qubits by 2 qubits, enabled by the GH200 Grace Hopper Superchip’s unified CPU-GPU memory architecture, per NVIDIA’s announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
When will the Ericsson 6G AI collaboration run on JUPITER?
NVIDIA’s official announcement does not list a specific start date for completed Ericsson 6G AI runs on JUPITER, only that the partnership was announced in March 2025 and will target large-scale AI model training for 6G network development, per NVIDIA’s ISC 2025 science showcase blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
Bottom line: Research teams with computationally intractable projects at the 1-exaflop scale can directly reference JUPITER’s validated, production-ready workflow templates: the Jülich Brain Atlas CytoNet model processes 6.5 petabytes of cellular brain imaging data from 21 post-mortem brains in under 5 days using 4,096 NVIDIA Grace Hopper Superchips, the ICON climate model simulates 146 days of fully coupled global Earth system dynamics at 1-kilometer resolution in 24 hours using 20,480 Grace Hopper Superchips, and the 50-qubit quantum simulation leverages unified CPU-GPU memory to bypass the 48-qubit ceiling of prior GPU-only systems. The announced Ericsson 6G AI collaboration, first publicized in March 2025, will add future telecom workload templates to JUPITER’s portfolio once runs complete, per NVIDIA’s official ISC 2025 science announcement blogs.nvidia.com/blog/jupiter-exascale-supercomputing-science/.
