NVIDIA: JUPITER — Europe's first exascale supercomputer sets scientific records at ISC 2026
JUPITER is Europe's first exascale supercomputer, located at Forschungszentrum Jülich in Germany. Powered by NVIDIA Grace Hopper Superchips and a Quantum-X800 InfiniBand network, JUPITER mapped 86 billion neurons, broke the quantum record by simulating 50 qubits, and set a world record in climate simulations.
This article was generated using artificial intelligence from primary sources.
JUPITER: Europe’s breakthrough in exascale computing
JUPITER (Joint Undertaking Pioneer for Innovative and Transformative Exascale Research) is located at Forschungszentrum Jülich in Germany and represents Europe’s first exascale supercomputer — a system capable of executing more than 10^18 arithmetic operations per second. The hardware foundation consists of up to 20,480 NVIDIA Grace Hopper Superchips interconnected by a Quantum-X800 InfiniBand network, while smaller configurations start from 4,096 chips depending on the workload type. The achievements presented at the ISC 2026 conference confirm that JUPITER is not merely a symbolic milestone for Europe but a system that changes what is even possible to compute.
What does JUPITER achieve where previous systems could not?
The answer lies in a combination of memory capacity and network bandwidth that enables computations previously unachievable within reasonable timeframes. The CytoNet foundation model, specialized in brain tissue analysis, processed 6.5 petabytes of data collected from 21 post-mortem brains in under 5 days, mapping all 86 billion neurons. The same task on a conventional cluster would have taken weeks. In parallel, an Earth system climate simulation at 1-kilometer resolution achieved approximately 146 simulated days of climate within 24 hours of compute time — a world record for this type of model. Previous generations of European systems operated at resolutions of 5 to 10 km, meaning precision increased 25 to 100 times.
Quantum record and industrial applications
JUPITER executed a full simulation of a 50-qubit quantum computer, surpassing the previous record of 48 qubits. A difference of two qubits sounds modest, but each additional qubit doubles the memory requirement of the simulation — simulating 50 qubits requires four times more resources than simulating 48. Beyond academic applications, Ericsson and Forschungszentrum Jülich are jointly using JUPITER to train AI models for future 6G networks, confirming that European exascale capacity is attracting the private sector as well. The NVIDIA Grace Hopper Superchip — a combination of an ARM-based Grace CPU core and Hopper GPU architecture on a single package — is key to the system’s energy efficiency compared to classic x86 plus discrete GPU configurations.
What this means for European computing independence
JUPITER operates within the EuroHPC Joint Undertaking program, whose goal is to reduce Europe’s dependence on infrastructure outside the EU. With this system, European researchers gain access to exascale capacity without needing to reserve time on American or Japanese systems such as Frontier (Oak Ridge) or Fugaku. The ISC 2026 results of JUPITER clearly position Europe as an equal participant in the global race for supercomputing dominance.
Frequently Asked Questions
- What does exascale mean and why is JUPITER important?
- Exascale refers to computing systems capable of more than 10^18 (one quintillion) operations per second. JUPITER is the first European system of this class and is already breaking world records in climate, neuroscience, and quantum simulations.
- What are the concrete results of the JUPITER supercomputer shown at ISC 2026?
- CytoNet processed 6.5 petabytes of brain data in 5 days, the climate simulation reached 1 km resolution for the entire Earth, and the quantum simulation surpassed the previous record of 48 qubits with a full simulation of a 50-qubit system.
Related news
NVIDIA: CUDA-X libraries cuPhoton, DAQIRI, and ALCHEMI accelerate astronomy, chemistry, and materials science
NVIDIA: Vera CPU at Los Alamos — 7× faster agentic AI for nuclear science and 3 new supercomputers
AMD: ROCm Optimization of Matrix3D for 3D Worlds Speeds Up Rendering by up to 54 Percent on Instinct GPUs