Nvidia Unveils Europe’s First Exascale Supercomputer

Based on Eviden’s BullSequana XH3000 liquid-cooled architecture, JUPITER can run 1 quintillion FP64 operations per second and is expected to offer up to 90 exaflops of AI performance.

  www.nvidia.com

• Climate science, including weather prediction and simulation
• Generative AI and large language models (LLMs) for European languages
• Neuroscience, including for drug discovery and mapping the human brain
• Quantum simulation, to bring quantum computing closer to reality
• Along with many other disciplines

For example, Max Planck Institute for Meteorology is using JUPITER to simulate climate predictions with a spatial resolution of about 1 kilometer, depicting extreme weather events such as violent thunderstorms and heavy rainfall much more realistically than before.




The Jülich Supercomputing Centre and a German consortium of nine European partners from research, industry and the media are tapping JUPITER for TrustLLM, a project that’s training the next generation of LLMs for various European languages. Such LLMs can ease workflows across virtually every industry by generating humanlike responses and improving productivity.

Neuroscience researcher Thorsten Hater at the Jülich Supercomputing Centre plans to use JUPITER to simulate the behavior of individual neurons on the subcellular level with the Arbor simulator. Such simulations will be crucial for developing therapies to combat neurodegenerative diseases like Alzheimer’s.

In addition, JUPITER is poised to break the world record for a supercomputer’s ability to handle qubits, the basic units of information in quantum computing. A typical laptop’s memory can handle about 32 qubits. The current record on a supercomputer is 48 qubits. An exascale computer like JUPITER could surpass 50 qubits — a significant milestone for quantum simulation.

In Germany and beyond, JUPITER is powering Europe’s most challenging research initiatives, helping scale the continent’s AI leadership with optimal energy efficiency.

Early flagship projects include:

• Molecular dynamics simulation: The Max Planck Institute of Biophysics will use JUPITER to simulate the nuclear pore complex — the largest protein assembly in cells — to achieve atom-level insights, advance nuclear transport models and combat retroviruses like HIV.
• Multilingual LLMs: With JUPITER, the University of Edinburgh will generate synthetic data to train LLMs that can reason over long documents in any language. More LLMs are being trained through the JUPITER Research and Early Access Program across Europe.
• Particle physics: The University of Wuppertal will use JUPITER to significantly increase the resolution of its microphysical computations, including to study the magnetic moment of an elementary particle, called muon, and potentially discover new particles and interactions.
• Foundation models for video: Using JUPITER, the Ludwig Maximilian University of Munich will develop spatio-temporal compression and diffusion architectures that enable the creation of high-quality, accessible video models to advance applications from medical imaging to autonomous driving.
• Multimodal AI foundation models: The University of Lisbon will tap JUPITER to scale multimodal, multilingual, open language models — integrating concepts from machine learning, sparse modeling, information theory and cognitive science — so these LLMs can support all European languages and address the limitations of existing models.

www.nvidia.com

  Ask For More Information…