The Jülich Research Center in North Rhine-Westphalia is working on Europe’s first exascale supercomputer. According to NVIDIA, JUPITER is set to become the most powerful AI system in the world.
• The first exascale supercomputer in Europe is being built in Jülich
• Booster module is based on NVIDIAS next generation Grace Hopper™ platform
• JUPITER is set to become the most powerful AI system in the world
Supercomputer in Jülich
As the Forschungszentrum Jülich announced at the beginning of October, the European supercomputing initiative EuroHPC JU and the supercomputer consortium ParTec-Eviden have signed the contract to build Europe’s first supercomputer of the exascale class called “Joint Undertaking Pioneer for Innovative and Transformative Exascale Research”, for short JUPITER, signed. The supercomputer will be stationed at the Forschungszentrum Jülich in North Rhine-Westphalia and “is designed for simulations and AI applications in science and industry that require maximum computing power,” as the Forschungszentrum Jülich press release states.
JUPITER should be used, among other things, in “training large neural networks such as AI language models” and in “simulations for the development of functional materials”. Other areas of application include “the development of digital twins of the human heart or brain for medical purposes” and “the validation of quantum computers and high-resolution climate simulations that cover the entire Earth system.”
Modular supercomputer architecture
According to the Jülich Research Center, JUPITER will be “based on a modular supercomputer architecture” and “will have a highly scalable booster module as well as a closely linked, universally applicable cluster module.” While the cluster module will be equipped with SiPearl’s new Rhea processor, the booster module, as NVIDIA recently announced, will be “based on the NVIDIA Grace Hopper™ accelerated computing architecture […]”to provide extremely large computing power for AI and simulation workloads”.
“JUPITER marks the debut of a quad-NVIDIA GH200 Grace Hopper superchip node configuration based on the liquid-cooled BullSequana connected,” writes NVIDIA in its press release dated November 13th. “The Quad-GH200 features an innovative node architecture with 288 Arm Neoverse cores capable of achieving 16 petaflops of AI performance with up to 2.3 terabytes of high-speed memory.” Four GH200 processors will be located in one of the liquid-cooled BullSequana XH3000s and will be networked via the high-speed NVIDIA NVLink® connection.
NVIDIA introduced its next-generation GH200 Grace Hopper™ platform, “based on a new Grace Hopper superchip with the world’s first HBM3e processor,” in August.
Most powerful AI system in the world
The parties involved see the new supercomputer as a significant development and great potential. “The JUPITER supercomputer, powered by NVIDIA GH200 and leveraging our advanced AI software, will deliver exascale AI and HPC power to solve the greatest scientific challenges of our time,” said Ian Buck, vice president of Hyperscale and HPC at NVIDIA, quoted in the company’s press release. “Our work with Jülich, Eviden and ParTec on this groundbreaking system will usher in a new era of AI supercomputing to advance the frontiers of science and technology.”
Thomas Lippert, director of the Jülich Supercomputing Center, has words of praise for NVIDIA: “The heart of JUPITER is NVIDIA’s accelerated computing platform, making it a groundbreaking system that will revolutionize scientific research.”
Emmanuel Le Roux, group senior vice president and global head of HPC, AI and Quantum at Eviden, said JUPITER is “the latest example” of the significant advances Eviden is making with the NVIDIA GH200.
According to NVIDIA, JUPITER is expected to be the most powerful AI system in the world and “over 90 exaflops of performance for AI training […] and 1 exaflop for “high performance computing (HPC) applications”. The supercomputer should only consume 18.2 megawatts of electricity.
The system and its operation over six years are expected to cost 500 million euros.
Installation of the JUPITER system is expected in 2024.
Editorial team finanzen.net
Selected leverage products on NVIDIA
With knock-outs, speculative investors can participate disproportionately in price movements. Simply select the lever you want and we will show you suitable open-end products on NVIDIA
The leverage must be between 2 and 20
Advertising
Image sources: Katherine Welles / Shutterstock.com, Andrey Suslov / Shutterstock.com