Europe has entered the big leagues of supercomputing with Jupiter, which this month became the first European system to achieve the exascale threshold of more than one quintillion (a billion billion) operations per second. Jupiter is Europe’s most powerful compute platform and the fourth fastest worldwide. It is a hybrid platform that uses a combination of SiPearl and Nvidia chips, respectively supporting HPC tasks like simulations and data analysis as well as AI workloads, such as training large language models and providing access to the Jupiter AI Factory (JAIF), a managed interface for developers and academics.

Based at the Jülich Supercomputing Center, Jupiter represents a $585 million joint investment by the European Union and Germany administered through the High-Performance Computing Joint Undertaking (EuroHPC JU),  according to the EU Commission news post.

“Jupiter was built by a consortium consisting of Eviden, a subsidiary of French tech giant Atos, and German group ParTec,” writes Tech Xplore, which says it is “the first supercomputer that could be considered internationally competitive for training AI models in Europe, which has lagged behind the United States and China in the sector.”

The Jupiter booster partition, an AI acceleration system built by Eviden, “integrates 24,000 Nvidia GH200 Grace Hopper Superchips interconnected with Nvidia Quantum-2 InfiniBand networking optimized for highly parallel applications such as training AI models or numerically demanding simulations,” according to an Eviden announcement.

France’s SiPearl developed the Rhea processor, an Arm-based system-on-chip (SoC) that Jupiter will rely on for traditional HPC workloads, such as climate modeling, quantum simulations, and computational biology, as detailed in a Jülich Supercomputing Center technological overview.

The system runs entirely on renewable energy, according to the EC, which says it is “the world’s most energy-efficient supercomputer module,” ranked number one on the Green500.

“Jupiter joins existing supercomputers in the EuroHPC network — namely, MareNostrum in Spain, Leonardo in Italy, Lumi in Finland, Discoverer in Bulgaria, MeluXina in Luxembourg, Vega in Slovenia, Karolina in Czechia and Deucalion in Portugal,” reports Silicon Republic, noting “the supercomputers are a part of Europe’s wider strategy to develop AI gigafactories.”

The European Commission describes the gigafactories as “large-scale computing hubs dedicated to training and deploying frontier AI models” and says the EuroHPC has so far selected 13 proposals for AI factories in France, Germany, Poland and Spain. “The factories will provide access to the massive computing power that startups, industry and researchers need to develop their AI models,” Silicon Republic writes.