New supercomputer for simulation and AI

[Translate to Englisch:] Symbolbild

The Erlangen National High Performance Computing Center based at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) has placed an order for two new supercomputers to be installed. The KU participates in these high-performance computers with four nodes. In the future, these computer capacities will primarily be used by the Mathematical Institute for Machine Learning and Data Science (MIDS) – among other things for research into weather and climate models.

At the end of December, the National High Performance Computing Center placed the order for the installation of a high-performance computer with more than 300 nodes, each with two AMD processors of the brand new "Turin" type with up to 2.3 terabytes of main memory. The computing nodes are linked via a high-speed network so that communication does not become a bottleneck during complex simulation calculations in view of the considerable computing power. The supercomputer is designed to deliver good performance and scalability for a wide range of applications from all scientific fields. It is expected to be installed in the first half of 2025. Another supercomputer will contain more than 40 nodes, each with four Nvidia H100 graphics/AI processors. It is tailor-made for machine learning and artificial intelligence applications. The installation of this "AI computer" is planned for the second half of 2024. A file system of more than 3000 terabytes offers sufficient capacity for data storage. It is connected to both clusters at high speed.

Both supercomputers will be available to scientists from all over Germany. Funding is provided by the Association for National High Performance Computing (NHR) and by the Bavarian Ministry of Science, FAU, KU and Hof University of Applied Sciences. The KU is contributing around 150,000 euros to the development – these funds come from the High-Tech Agenda and are part of the equipment for the KU Chair of Reliable Machine Learning. The integration of the KU nodes into the new NHR clusters guarantees a professional setup and support and allows load peaks to be balanced out. There is also the possibility of increasing the quotas for large research projects.

Contract signing
Dr. Nils Blümer (KU, from left), Prof. Dr. Michael Hartmann (FAU), Prof. Dr. René Peinl (HS Hof), André Singer (MEGWARE GmbH) and Prof. Dr. Gerhard Wellein (FAU) on the day the contract was signed. Also present was the founder of FAU, Margrave Friedrich von Brandenburg-Bayreuth (front).

The Mathematical Institute for Machine Learning and Data Science (MIDS) at the KU will use the new supercomputers for various research projects. For example, the team led by Prof. Dr. Tijana Janjic, Professor of Data Assimilation, is working on the question of how weather forecasts can be improved by developing algorithms that combine physical knowledge and data. Prof. Dr. Götz Pfander's Chair of Mathematics and Scientific Computing will use the high-performance computing capacity to create and verify hypotheses before proving them mathematically. The chair is working on the generation and mathematical analysis of complex exponential functions. In practice, the findings are used in areas such as signal processing in mobile communications.

Prof. Dr. Nadja Ray's Chair of Geomatics and Geomathematics uses the high-performance computer to simulate the transport of pollutants such as radioactive substances, heavy metals or microplastics in soils – which also requires a great deal of computing power. The research groups led by Prof. Dr. Felix Voigtlaender, Chair of Reliable Machine Learning, and Junior Professor Dr. Dominik Stöger (Data Science) will carry out mathematical analyses of machine learning algorithms on the cluster. The two mathematicians are interested in developing and testing hypotheses in the field of machine learning. "With pen and paper", they say, it was often very time-consuming to make assumptions and verify whether they are correct. The cluster thus saves valuable time, which can be used to prove the empirically found properties.