New Supercomputer Accelerating Top-Level Research

Constantly increasing data volumes, ever more complex calculation and modeling – working with large amounts of data is crucial in many fields of research. The new high-performance computer “Marvin” now makes it easier for University of Bonn researchers to leverage artificial intelligence and machine learning. Featuring a scalable GPU partition, Marvin is among the world’s top 500 most powerful computers. Funded in part under the Excellence program, the acquisition of Marvin comes as part of the University of Bonn digital strategy.

As Marvin’s 18,432 computing cores commenced their work at the symbolic push of a button, a new era of powerful, high-performance computing (HPC), artificial intelligence (AI) and machine learning dawned for University of Bonn researchers. High-performance computing will now be a driver behind AI-informed solutions for more sustainable agriculture, evolving or even revolutionary battery technology through computational chemistry and more accurate climate research and modeling, to name just a few applications.

“The startup of mainframe computer Marvin represents a milestone in high-performance computing at the University of Bonn, as a core element within our forward-looking strategy. Marvin creates tremendous opportunities for our researchers, whose work involves ever-larger and increasingly complex datasets. These new resources and computing capacity mean we are excellently positioned within the competitive international environment,” commented Rector Michael Hoch.

New possibilities for research involving huge data volumes

Marvin’s ability to process and analyze vast amounts of data opens up new possibilities for research, in such areas as robotics, climate research, astrophysics, life sciences, sustainable crop production … and even the humanities. “Marvin will allow interdisciplinary collaboration on complex projects among researchers from different fields, enabling answers to be found to current research problems and questions in the form of innovative, jointly derived solutions,” said Professor Maren Bennewitz, University of Bonn Vice Rector for Digitalization, who was instrumental in the acquisition of the mainframe to realize a major milestone within the University’s digital strategy.

“Marvin’s ample capacity means researchers can conduct top-level research, enjoying easy access right here on campus with scientific support from the interdisciplinary team at the HPC/A-Lab,” said Professor Mutzel, referring to the High Performance Computing and Analytics Lab, which administers all HPC activities at the University of Bonn. The Lab’s mission includes planning and coordination of the central high-performance computing infrastructure of the University of Bonn in collaboration with University IT. The HPC/A-Lab provides advice enabling researchers to swiftly resolve HPC-related issues on their own in a resource-efficient manner. Interdisciplinary projects like “BNTrAinee” are also underway, aimed at transporting existing AI expertise out to different faculties and departments in order to promote the collaborative development of needs-based teaching and study offerings.

The HPC team newly formed within University IT provides technical and general support and oversees operation. “Rolling out this supercomputer at the University of Bonn is definitely a major milestone, and we are excited to now offer highly sophisticated scientific computing services, thanks to Marvin,” said University IT Director Dr. Rainer Bockholt. Offering high-performance computing cores and graphics processors (GPUs), Marvin is available for use by all University of Bonn departments and researchers at our partner institutions (exclusively).

Research use cases

Marvin is used by Bonn-based climate researchers to integrate satellite data within numerical models of the Earth system. “This helps us improve modeling, enabling more accurate prediction of droughts and sea level changes,” notes Professor Jürgen Kusche of the Institute for Geodesy and Geoinformation. That includes gaining an understanding of the sea-level impact of melting of the Greenland ice sheet, as well as how large-scale droughts arise and the effects they have – as he further elucidated. “Thanks to Marvin, tweaking our modeling takes less time, and we enjoy much greater flexibility,” Professor Kusche observes.

The concentrated computing power of Marvin will also be used within the PhenoRob Cluster of Excellence, In the area of sustainable agriculture, for example, sensor data can flow into AI algorithms to run simulations to enable targeted crops fertilization or more sustainable weed control. “The generous GPU resources significantly accelerate our computing-intensive learning algorithms, and Marvin affords us optimizations of even bigger models,” reports Cluster spokesperson Professor Cyril Stachniss.

The high-performance computer will be used for theoretical chemistry at the University of Bonn as well, in battery research, for example, where high-performance computing is crucial for innovation. In computational chemistry, the complex behaviors and interactions of great numbers of molecules are calculated at the transition point from liquid to solid, yielding an understanding of the molecular properties within a larger, overarching system. Such basic research has enabled the identification of suitable electrolytes for the batteries of the future. “Here in Bonn part of our interest in using new computing methods is to derive effective solutions for such complicated applications,” commented Professor Barbara Kirchner of the Mulliken Center for Theoretical Chemistry. “That includes suitable solvent mixtures (solutions in the chemical sense of the word) that serve to meet many requirements, like sustainability, in differing situations.”

Technical setup, “Top 500” ranking

Supercomputer Marvin has two primary components which make it a powerful and versatile tool for research and development, the first of these being its set of 192 dual-socket computing nodes with 384 Intel Xeon Sapphire Rapids processors, optimized for massively parallel processing. Each of these processors has 48 cores, giving Marvin an impressive total of 18,432 computing cores.
Marvin’s second component is a set of 32 nodes, each equipped with four Nvidia A100 GPUs, which enables the use of highly scalable algorithms for simulations and data analysis. Marvin also has another set of 24 nodes consisting respectively of eight Nvidia A40 GPUs, specially designed for machine learning applications; this component will be useful in supporting a wide range of projects.
Marvin’s scalable GPU partition ranked 423rd in the latest Top 500 ranking of supercomputers worldwide. This component furthermore placed 34th in the Green500 ranking for efficiency. The mainframe therefore runs with theoretical maximum performance of 2.63 petaflops (quadrillion floating point operations per second).

Media contacts:

For scientific/research-related questions, Petra Mutzel (HPC/A):
Director’s Office of the High Performance Computing and Analytics Lab
Phone: +49 228 73-69917
Email: petra.mutzel@cs.uni-bonn.de

For technical inquiries, Dirk Barbi (HPC/University IT):
HPC Team Leader
Phone: +49 228 73-66111
Email: dbarbi@uni-bonn.de

Weitere Informationen:

https://www.dice.uni-bonn.de/hpca/de
https://www.hpc.uni-bonn.de/en