MIL OSI Translation. Region: Germany / Deutschland –
Source: Forschungszentrum Julich Forschungszentrum Jülich and its partners have designed an incomparably flexible and energy-efficient supercomputer – it is the fastest computer in Europe.
Jülich, November 16, 2020 – The Jülich supercomputer JUWELS has been expanded in the last few months. Thanks to a new booster module, 85 petaflops are now possible, which corresponds to 85 quadrillion computing operations per second or the computing power of more than 300,000 modern PCs. JUWELS can massively expand the limits of simulations and also offers the strongest platform in Europe for the use of artificial intelligence (AI). The computer that Forschungszentrum Jülich, the Franco-German company Atos and the Munich supercomputing specialist ParTec have developed together with the US manufacturer NVIDIA is currently the fastest system in Europe. The Jülich supercomputer, which is run by the national Gauss Center for Supercomputing funded reached 7th place on today’s TOP500 list of the fastest computers in the world. JUWELS ranks third on the current Green500 list and is the most energy-efficient system in the highest performance class.
Video: Flexible supercomputer made in Jülich (length: 4:01 min.)
“We see supercomputing not only as an object of our research, but above all as a powerful tool with which we can answer complex research questions together with our partners from science and industry,” says Prof. Wolfgang Marquardt, CEO of Forschungszentrum Jülich With a fully expanded JUWELS system, Forschungszentrum Jülich gives scientists from a wide variety of institutions and scientific disciplines access to high-performance computing capacities at the very highest level. At the same time, however, with the system we are also demonstrating the responsible handling of the ever increasing energy demand for the provision of computing power. “Greater realism” A very topical example in the current COVID-19 crisis provides support for drug development on the computer, “explains Prof Thomas Lippert, the head of the Jülich Supercomputing Center. “Only the computing power of the booster enables our researchers to simulate the processes before, during and after a potential active ingredient meets a receptor or protein in a realistic way.” Active ingredient research on biologically relevant time scalesCopyright: Forschungszentrum Jülich, HHU Düsseldorf / C. Pfleger, H. Gohlke Another example is the detailed simulation of surface, earth and groundwater movements. With the new JUWELS booster, researchers are for the first time able to carry out simulations for Germany and Europe with the required fine resolution, for example of individual slopes or river corridors. Intelligent division of tasks – maximum energy efficiencyJUWELS is based on a highly flexible modular architecture jointly shared by Forschungszentrum Jülich has developed with European and international partners. “With its powerful, highly efficient graphics processors, the new booster module is specially designed for extremely computationally intensive applications that can easily be processed in parallel on a large number of computing cores,” explains Dr. Dorian Krause, who is responsible for the construction and operation of the extraordinarily complex system in Jülich. “In addition, JUWELS is among the top 10 fastest computers in the world, leading the way in energy efficiency.” NVIDIA A100 Tensor Core GPU Copyright: Forschungszentrum Jülich / TRICKLABORD The Jülich supercomputer is one of the first to be equipped with NVIDIA A100 Tensor Core GPUs based on the NVIDIA Ampere architecture based. The booster combines around 12 million so-called CUDA cores (FP64) on its more than 3,700 graphics processors, which are connected to each other via an NVIDIA Mellanox HDR Infiniband high-performance network with 200 Gb / s. The booster alone achieves a peak performance of 73 petaflops. Especially for AI applications that place different demands on the hardware, even up to 2.5 exaflops are possible: this corresponds to 2.5 trillion arithmetic operations per second. The module is thus the strongest platform in Europe for the use of artificial intelligence (AI). “The highlight of JUWELS is that both modules, the previous ‘cluster module’, which works with fast processors (CPUs), and the booster module with its GPUs, are very closely interconnected, ”says Bernhard Frohwitter, CEO of the Munich supercomputing specialist ParTec. The cooperation between the modules controls ParTec’s modular software system ParaStation Modulo, a leading global development from Germany. “With ParaStation Modulo, JUWELS can dynamically access CPUs and GPUs within a code and optimize the calculation.” Water cooling in the JUWELS supercomputerCopyright: Forschungszentrum Jülich / TRICKLABOR water-cooled, patented DLC (Direct Liquid Cooling) solution contributes significantly to the low energy consumption of the system, ”explains Agnès Boudot, Senior Vice President, Head of HPC & Quantum at Atos. “The Atos design ensures that the computing power of the CPU and GPU blades can be fully used by applications.” Prepared for future technologiesFor Prof. Thomas Lippert, the JUWELS system is a milestone towards the European exascale computer, the to start from 2023. The construction and operation of such a supercomputer is considered the next big step in supercomputing worldwide. With a computing power of at least one exaflops, i.e. 1 trillion floating point operations per second, it would be at least twelve times faster than the JUWELS supercomputer. “JUWELS modular architecture, the design of its computing nodes, the network, the infrastructure and the cooling as well as the software Architecture can easily be transferred to an exascale computer, whereby costs and energy expenditure remain acceptable, ”explains Thomas Lippert. JUWELS is also perfectly prepared for the future of supercomputing in other respects. The modular design makes it possible to integrate future technologies that are also being intensively researched at Forschungszentrum Jülich: These include, for example, quantum computer modules or neuromorphic modules that work on the model of the human brain. JUWELS supercomputers in the computer hall of the Jülich Supercomputing Center Forschungszentrum Jülich / Wilhelm-Peter Schneider Funding by the federal and state governments The acquisition of the JUWELS booster is financed by the federal government and the state of North Rhine-Westphalia. The JSC operates JUWELS as a member of the Gauss Center for Supercomputing (GCS), the amalgamation of the national high-performance computing centers in Germany, which includes the three computing centers of the Forschungszentrum Jülich (JSC), the Bavarian Academy of Sciences (LRZ) and the University of Stuttgart (HLRS) The computing time is allocated after the application and scientific assessment at national and European level. The GCS and Forschungszentrum Jülich are supported by the Federal Ministry of Education and Research (BMBF) and the Ministry of Culture and Science of the State of North Rhine-Westphalia as well as the Ministry of Science, Research and the Arts Baden-Württemberg and the Bavarian State Ministry for Science and Art.
The modular concept implemented at the Jülich Supercomputing Center (JSC) was developed over many years of cooperation with the Munich software company ParTec. It provides a supercomputer made up of several specialized components that can be dynamically combined as required using standardized software. Since 2011, European partners from industry and research have developed and tested the first modular systems under Jülich management, and have continuously expanded the concept in the EU-funded DEEP research projects. Manufacturer Atos (France), the software specialist ParTec (Germany) and the graphics processor manufacturer NVIDIA (USA). The Israeli network specialist Mellanox, also involved, was taken over by NVIDIA in the spring of this year.
JUWELS-ClusterJUWELS-BoosterComputing power12 petaflops (12 quadrillion arithmetic operations per second) 73 petaflops (73 quadrillion arithmetic operations per second) arithmetic nodes2511 CPU nodes + 56 GPU nodes936 GPU nodesprocessors total 5134 CPUs (Intel Xeon Skylake) + a total of 224 GPUs (NVIDIA V100s) 1872 CPUs (AMD EPYC Rome) + a total of 3744 GPUs (NVIDIA A100) processing cores 122,768 CPU cores + 71,680 FP64-CUDA cores (GPUs in total) 44,928 CPU cores + 12,939,264 FP64-CUDA cores (GPUs in total) RAM in total 275 TB total 479 TB + total 150 TB high bandwidth memory network 100 Gb / s (Mellanox InfiniBand EDR) 200 Gb / s (NVIDIA Mellanox HDR InfiniBand)
Early Access Applications: The first applications
Computing time on the JUWELS booster is precious. The first applications were already running on the system during the first test phase in late summer and autumn of this year. The so-called “Early Access Applications” provide valuable information for optimizing the code and provide the first tangible scientific results.
Supercomputer for big arithmetic tasks and AI
Thanks to a new booster module, JUWELS has increased its computing power enormously. In recent months, researchers from all over Europe have adapted and developed programs so that they run massively in parallel on the more than 3,700 graphics processors (GPUs) of the booster. Read more …
Further information: Press release for download (pdf) Press release from November 14, 2019, A turbocharger for the JUWELS Jülich Supercomputing CentreTOP500 list of the fastest computers in the world Images: Images may be used for editorial reporting provided the source is acknowledged. Copyright: Forschungszentrum Jülich / Wilhelm- Peter SchneiderCopyright: Research Center Jülich / Wilhelm-Peter SchneiderCopyright: Research Center Jülich / Wilhelm-Peter SchneiderCopyright: Research Center Jülich / TRICKLABOR contact: Dr. Dr. Thomas LippertDirector of Jülich Supercomputing CentreTel .: +49 2461 61-6402E-Mail: firstname.lastname@example.orgDr. Dorian KrauseJülich Supercomputing CentreTel .: 02461 61-3631E-Mail: email@example.comPress contact: Tobias Schlößer Corporate CommunicationsTelephone: +49 2461 61-4771E-Mail: firstname.lastname@example.org
EDITOR’S NOTE: This article is a translation. Apologies should the grammar and / or sentence structure not be perfect.