The JUPITER Supercomputer in Germany: Potential to Become the Fastest Supercomputer in the World for AI Applications.
Europe is gearing up to build the first exascale supercomputer named JUPITER, which could become the fastest and most powerful machine in the world, according to a report by Interesting Engineering on December 29. JUPITER is set to pave the way for unprecedented scientific discoveries and innovations in fields ranging from climate change to quantum physics by performing calculations that very few computers in the U.S. and China can currently execute.
Jülich Supercomputing Centre, where JUPITER is installed. (Photo: Jülich Supercomputing Centre).
This supercomputer aims to achieve groundbreaking results in Europe by conducting one quintillion arithmetic operations per second. This exceptional computing power will provide new directions for scientific simulations and drive revolutionary advancements in the field of artificial intelligence.
The exascale supercomputer can perform one quintillion calculations per second, equivalent to the computing power of 10,000 laptops combined. Currently, there are only two officially recognized exascale supercomputers in the world: Frontier at Oak Ridge National Laboratory in Tennessee and Aurora at Argonne National Laboratory in Illinois. However, China may have at least two untested exascale machines that are not yet ranked on the list of the 500 most powerful supercomputers in the world.
JUPITER (Joint Undertaking Pioneer for Innovative and Transformative Exascale Research) will be constructed at the Jülich Supercomputing Centre in Germany by the European High-Performance Computing Joint Undertaking (EuroHPC JU), a collaborative project between the European Union and private companies. In a recent interview, Professor Thomas Lippert, director of the Jülich Supercomputing Centre, stated that JUPITER has the potential to become the fastest supercomputer in the world for AI applications, with over 90 exaflops at 8 bits. An exaflop is a unit of measurement for the overall computing power of computer systems, with one exaflop equivalent to one quintillion calculations per second.
JUPITER will feature a unique modular design consisting of two main components: the Booster Module and the Cluster Module. The Booster Module will utilize advanced technology from NVIDIA (with nearly 24,000 NVIDIA GH200 GPUs) to provide computing power for AI applications and simulations, such as training AI models capable of generating realistic images and text.
The Cluster Module will employ the new Rhea processors from SiPearl, produced in Europe, to tackle complex tasks that require high memory bandwidth. The two modules will be integrated by Eviden into their energy-efficient and liquid-cooled BullSequana XH3000 platform, operating as a coherent supercomputer under the ParaStation Modulo operating system from ParTec. JUPITER is the result of collaboration between ParTec, Eviden, SiPearl, and NVIDIA with the European scientific community. It will be the first exascale supercomputer in Europe.
The installation process for JUPITER is set to begin in early 2024. Users will have the opportunity to experiment and prepare for the system through the JUPITER Early Access program, allowing for close collaboration among stakeholders to develop and optimize the best possible version of the system for the scientific community.
Philippe Notton, CEO of SiPearl, stated that the machine has the potential to top the list of the 500 supercomputers if there are no faster competitors. According to Notton, JUPITER’s final ranking will depend on the performance of the institutions operating the machine and hardware evaluation metrics. However, he also emphasized the importance of reducing energy consumption, as the operational cost of running JUPITER over three years is estimated to be around $553 million, including an estimated $110 million in annual electricity costs.
