---

---

24.04.2024 10:25

Innovation booster at TU Dresden: A network of three new supercomputers for research inaugurated

Jacqueline Papperitz Pressestelle
Technische Universität Dresden

The Center for Information Services and High-Performance Computing (ZIH) at TUD Dresden University of Technology (TUD) inaugurated the highly specialized new "Barnard" supercomputing system and the first components of the "SpiNNaker2" computer architecture at a ceremony held on 23 April 2024. At the same time, the procurement of another computing cluster, "Capella", was also presented.

As an integrated system, the trio addresses the requirements for high computing power in close connection with large amounts of data as well as special solutions for artificial intelligence and machine learning, thus meeting the current needs of research. As a key technology for innovative developments in a variety of scientific fields, the systems within the NHR (National High-Performance Computing) network offer excellent conditions for breaking new ground competitively and implementing rapid solutions for artificial intelligence.

While Barnard is a CPU-based high-performance computing system (CPU: Central Processing Unit), particularly for data-intensive requirements, the current procurement of Capella complements a strategically significant increase in computing capacity for machine learning and will also accelerate many other scientific applications in simulations and data analytics that are optimized for the GPU architecture (GPU: Graphics Processing Unit). The ZIH consistently uses hot water cooling for its systems in high-performance computing, which achieves a high degree of cooling efficiency. The two systems are also integrated into the sustainable concept of waste heat utilization. In addition to heating the surrounding buildings, heat surpluses will also be fed into the district heating network in future. With its unique hybrid AI framework, the system design of SpiNNaker2 enables real-time AI on an unparalleled scale. The configuration of the system is characterized by high energy efficiency.

Many high-profile guests from politics, science and industry attended the inauguration ceremony. In their festive speeches, the Rector of TUD, Prof. Ursula Staudinger, and the Saxon State Secretary for Digital Administration and Administrative Modernization, Prof. Thomas Popp, underlined the importance of technologies such as supercomputing and AI for science and for Saxony as prime research location.

"TUD Dresden University of Technology is a lighthouse of digitalization competence in Germany. As one of nine centers of national high-performance computing, we continuously are pushing the boundaries of knowledge despite the growing intensity and complexity of data in the sciences," emphasized Prof. Staudinger. "The new supercomputing systems at TUD are a pivotal milestone for data-driven high-performance computing and the further development of artificial intelligence applications."

In an accompanying scientific colloquium, three scientists from the fields of materials science, astronomy and tumor research showcased their research, which is enabled or bolstered by the use of high-performance computing. In their presentations, they stressed the importance of supercomputing and AI for research in their fields. (Link to the colloquium: https://tu-dresden.de/zih/die-einrichtung/termine/zih_kolloquium/kolloquium-zur-...)

"Pressing current scientific issues rely on complex calculations, models and evaluations. Whether climate studies, medical research or engineering: the possibilities of the new system - technologies and infrastructure, software solutions, innovative methods and expertise - open up new dimensions for a variety of research areas," emphasized Prof. Wolfgang E. Nagel, Director of the ZIH, adding: "Providing the resources and addressing the associated challenges, such as sovereignty, innovative methods of artificial intelligence or even our own generative language models, are of strategic significance for Germany as a science location."

Presentation of the systems

Barnard

The Barnard supercomputing cluster by Eviden (https://eviden.com/de-de/) with fast Intel CPUs offers a total of more than 60,000 computing cores. Together with a fast InfiniBand interconnection network and a large hierarchical storage system (more than 40 petabytes), it enables the flexible provision of high computing power for data-intensive computing. This is essential, for example, for research into viruses or for analyzing astrometric data from the GAIA mission. It enables unprecedented insights into the structure and development of the Milky Way, supports fundamental research into innovative materials, more powerful battery storage, more efficient and environmentally friendly processes for the chemical industry and much more.

The system was largely funded by the National High-Performance Computing (NHR@TUD; in equal parts by the BMBF and the Free State of Saxony) and has been available for scientific use throughout Germany as part of the NHR network since fall 2023. The TUD's Physics of Life (PoL) Cluster of Excellence also contributed to the funding in order to secure dedicated computing power for its research.

Barnard stats

Flexible, high computing power for data-intensive high-performance computing

Peak performance - approx. 4.2 petaflop/s

Computing nodes - 630 nodes, each with 2 CPUs and 512 GB DDR5 main memory

Processors - A total of 1,260 Intel Xeon Platinum 8470 "Sapphire Rapids" (52 cores each)

Computing cores - 65,520 in total

Storage system
- 20 PB of fast SSD/HDD storage with 420 GB/s bandwidth,
- 20 PB additional HDD cache

Kosten - EUR 13 million

Availability - in operation since fall 2023

SpiNNaker2

The "Spiking Neural Network Architecture" SpiNNaker is a computer architecture that is optimized for the simulation of neural networks. It integrates 5 million ARM processor cores and various special co-processors in a system architecture optimized for communication and memory access. SpiNNAker2 is thus the largest system for biologically inspired artificial intelligence in the world. With operation at the minimum energy point of the processing units at 0.50 V and adaptive, fine-granular load control, SpiNNaker2 utilizes the innovative power-saving properties of Globalfoundries Dresden's flagship 22FDX technology. The neuromorphic computing system was jointly developed by TUD and the University of Manchester as part of the EU flagship project "Human Brain", with the participation of TUD spin-offs Racyics GmbH and SpiNNcloud Systems GmbH. In addition to the synergetic combination of various forms of AI and real-time capability, SpiNNaker2 also integrates quantum-inspired computation with randomness. This unique combination of functions enables the system to make innovative leaps in various fields of application, e.g. accelerating the search for active ingredients for personalized medicine by a factor of 10 to 100. SpiNNaker2 was driven by Prof. Christian Mayr, Chair of Highly-Parallel VLSI Systems and Neuromicroelectronics at TUD, and supported by ScaDS.AI Dresden/Leipzig. The first units are currently in trial operation, and once the installation is complete in summer 2024, the system should be available to a global user base from the fall.

The tapeout was financially supported by ERDF funding and the Free State of Saxony in the "SpiNNcloud" project (funding code: 100373652). (Link to the project: https://spinncloud.com/)

SpiNNaker2 stats

Cloud-based system for bio-inspired AI

Storage - 8.6 TB per rack (43 TB in total)

Number of neurons/synapses - 10 billion neurons, 10 bill. synapses

Processing cores - 5 million ARM processor cores

Costs - 9 million euros

Availability - Partly installed, completion in summer 2024

Capella

The Capella cluster, which was awarded to the Saxon company MEGWARE (https://www.megware.com/) at the end of 2023, will be equipped with more than 120 nodes, each with four H100 accelerators from NVIDIA and two AMD processors. In addition, a 1.1 PB storage system by the American software manufacturer WekaIO with high bandwidths of up to 1.9 TB/s will accelerate data-intensive applications as a "burst buffer". With the combination of more than 480 fast AI accelerators and the fast cache for providing the data, Capella is particularly geared towards applications in the field of artificial intelligence. For instance, Capella will make an important contribution to the training of European language models as part of the OpenGPT-X and Gaia-X projects or support medical research in the use of new machine learning methods - for example for cancer diagnosis and the development of new drugs.

The cluster will offer a wide range of options for measuring the energy requirements of components and support research into optimizing efficiency in a targeted manner. With the integration of Capella into the existing infrastructure in the data center, Barnard's file systems will also be efficiently connected in the overall system. The system will be available to scientists from all over Germany from the third quarter of 2024.

The cluster is funded by the National High-Performance Computing (NHR@TUD; in equal parts by the BMBF and the Free State of Saxony) and the AI Competence Center ScaDS.AI Dresden/Leipzig.

Capella stats

Computing capacity for machine learning, data analytics and HPC simulations

Peak performance - approx. 33 petaflop/s (FP64 tensor), approx. 0.98 exaflop/s (FP16 tensor)

Computing nodes - 124 with 2 CPUs, 4 GPUs and 768 GB DDR5 main memory each

Processors - 248 AMD Epyc 9334 "Genoa" (32 cores each)

GPU accelerators - 496 NVIDIA H100 with 94 GB HBM2e memory each

Storage system - 1.1 PB SSD storage with 1.9 TB/s bandwidth (read),
Connection to the Barnard storage systems

costs - EUR 12 million

Availability - User operation from the third quarter of 2024

Further information

About the ZIH
The department of the CIDS - Center for Interdiciplinary Digital Sciences, founded in 2021, operates the central IT infrastructure services and servers, including HPC systems. In addition, the ZIH carries out its own extensive research and development work and is closely networked nationally and internationally with scientific institutions and partners from industry and business. Since 2021, the ZIH has been offering access to dedicated supercomputing systems as well as the necessary expertise and support for their scientific use as a partner NHR@TUD in the existing NHR network (National High-Performance Computing). The aim of NHR is to provide researchers at German universities with the computing capacity they need for their scientific work, free of charge, to support them in using these resources with complex algorithms and AI methods, and to strengthen their skills in using the systems efficiently. Methods of machine learning and artificial intelligence, closely linked to the evaluation of large amounts of data, have long been part of the strategic orientation of the ZIH - be it with the establishment of the Competence Center ScaDS.AI Dresden/Leipzig founded for this purpose in close cooperation with the University of Leipzig or the successful application as an NHR center for the German science system. Link: https://tu-dresden.de/zih

Chair of Highly-Parallel VLSI Systems and Neuromicroelectronics
The Chair's core research areas are the design of systems-on-chip in deep-submicron (currently up to 22nm), sensor/actuator interfaces, signal processing and AI algorithms. Applications of the systems developed at the Chair lie in the simulation of brain functionality, multi-processor chips in an industrial context (telecommunications, smart city/smart farming, autonomous driving, Industry 4.0) as well as for AI accelerators and neuroimplants. The Chair currently has four spin-offs with 250 employees (Racyics, Coinbau, Siliconally, SpiNNcloud Systems). The spin-off Racyics is the largest independent pure-play chip design company in Europe and achieves world records, e.g. in ultra-low-energy microcontrollers. Christian Mayr is PI in the 5G Lab Germany, in the CeTi and cfaed Clusters of Excellence, in the 6G life Cluster, in the AI Competence Center ScaDS.AI Dresden/Leipzig and in various large-scale EU projects. An edge system from the Chair recently won first prize in the BMBF's AI hardware competition (1 million euros in prize money). Link: https://tu-dresden.de/ing/elektrotechnik/iee/hpsn

TUD Dresden University of Technology, as a University of Excellence, is one of the leading and most dynamic research institutions in Germany. With around 8,300 members of staff and around 29,000 students in 17 Faculties, it is one of Europe’s largest technically-oriented universities. Founded in 1828, today it is a globally oriented, regionally anchored top university, developing innovative solutions for the world's most pressing issues. In research and academic programs, the university unites the natural and engineering sciences with the humanities, social sciences and medicine. This wide range of disciplines is an outstanding feature that facilitates interdisciplinarity and transfer of science to society.

---

Wissenschaftliche Ansprechpartner:

For Barnard and Capella:
Prof. Wolfgang E. Nagel
Director of the Center for Information Services and High Performance Computing (ZIH)
TU Dresden
Center for Interdisciplinary Digital Sciences (CIDS)
wolfgang.nagel@tu-dresden.de

For SpiNNaker2:
Prof. Christian Mayr
TU Dresden
Chair of Highly-Parallel VLSI Systems and Neuro-Microelectronics
christian.mayr@tu-dresden.de

---

<
F.l.t.r: State Secretary Prof Thomas Popp, Prof Ursula Staudinger, Rector of TU Dresden, Prof Wolfga ...

Sven Geise/TUD

---

Merkmale dieser Pressemitteilung:
Journalisten, Wissenschaftler
Informationstechnik
überregional
Forschungsprojekte, Kooperationen
Englisch

---