25 Years of National High-Performance Computing at LRZ

The computing power of the supercomputers at LRZ increased approximately tenfold every four years, while storage capacity doubled after a little over a year. Similar developments have been observed at other data centres around the world: according to Moore's Law, the computing power of the systems ranked first in the Top500 list has doubled every 14 months. However, the top rankings of all LRZ supercomputers — from HLRB to SuperMUC to SuperMUC-NG — fade into the background when one considers the scientific achievements these supercomputers have enabled over the last 25 years. They have enabled manyfold “aha moments among scientists.

• Under the lead of scientists at the Technical University of Munich (TUM), the LRZ supercomputer produced a revolutionary model of the human lung, showing that it is not grape-shaped but sponge-like and larger than the skin when unfolded.
• Researchers at Ludwig-Maximilians-Universität Munich (LMU) used SuperMUC to simulate the formation of the Earth's continents over millions of years, as well as the processes occurring within their cores and crusts. The fascinating images resulting from this simulation are now on display at the American Museum of Natural History in New York.
• In 2013, the 1Kite project used SuperMUC to reconstruct the family tree of insects and birds. These fundamental biological findings remain among the most frequently cited results in the field.
• For the Millennium TNG astrophysics project, led by the Max Planck Institute for Astrophysics (MPA), LRZ's HPC systems simulated dark matter several times. This led to the discovery and measurement of neutrinos, among other things.

Meanwhile, the sound barrier was discovered in 2021 in the largest interstellar turbulence modelled and visualised by a team from Germany, the USA and Australia at LRZ. The repetition of this discovery in 2025 highlights the insatiable demand for computing power and storage space: whereas 23 terabytes of hard drive space were required for the previous simulation, the new one already occupies 30 terabytes due to its higher resolution. While 65,000 computing cores were previously in operation, 138,240 cores are now active. While the first simulation required around 130 terabytes of memory, the new one needs 288 terabytes. And researchers want to calculate even more.

Usage data from LRZ over the last 25 years clearly shows that simulation and modelling have become firmly established in the natural sciences, alongside experiments and measurements, as the third pillar. The classic heavy users in this field include not only astrophysicists, thermodynamicists, chemists and engineers, but also biologists, medical scientists, pharmacologists, climate researchers, mathematicians and computer scientists.

This versatility is the result not only of sophisticated computer technology, but also of active promotion by the Competence Network for Scientific High-Performance Computing (KONWIHR). Since its establishment in 2000, KONWIHR, which is funded by the Bavarian Ministry of Science, has been working to integrate HPC into a wider range of research disciplines in Bavaria. To this end, KONWIHR brings together expertise and HPC capabilities to support scientists in developing and optimising codes, and in maximising the potential of existing high-performance computers. 

In the coming years, artificial intelligence (AI) is set to support not only the Bavarian HPC community, but also facilitate an increase in the number of interdisciplinary projects, as high-performance computing and digitalisation continue to connect scientific disciplines.

Technically speaking, the end of Dennard scaling has shaped the last few years in HPC, and Moore's Law is also reaching its limits. Consequently, it is no longer possible to increase computing power and storage space simply by densifying transistors further or parallelising more processors. This is especially true given that the energy requirements of supercomputers are gradually exceeding the capabilities of data centres. Even the Top500 list only reflects the performance of the systems it compares to a limited extent. In light of these challenges, high-performance computing is driving research and development. At LRZ research continues on creating the best system for science rather than the fastest system as well as on energy-efficient operations of high-performance IT infrastructures.

Accordingly, the LRZ developed hot water cooling for SuperMUC in collaboration with technology partner IBM, launching it in 2012, and has since worked with Lenovo to continuously optimise it. With the expansion phase of SuperMUC Next Generation (SuperMUC-NG for short), LRZ hosts its first supercomputer that cools exclusively with hot water and uses waste heat for heating is now in place. Water cooling has become an HPC standard worldwide in supercomputing and helps data centres significantly reduce their energy consumption. 

The future of HPC at LRZ is also becoming clearer: the expansion of the current supercomputer, SuperMUC-NG, includes almost 27,000 CPUs and around 1,000 GPUs, which accelerate computing and enrich HPC and classic simulations with AI methods.

SuperMUC-NG, the LRZ's most powerful computer, has also been connected to quantum computers, opening up new possibilities for calculations. In summer 2025, the LRZ installed its first photonic processors. These processors use light instead of electrical impulses for computing and are currently undergoing testing to determine their compatibility with high-performance computers in an analogue-digital architecture.