DaREXA-F

Future exascale systems will have unprecedented computing power with over 10^18 floating point operations per second, which will be provided on thousands of heterogeneous compute nodes (probably with CPUs and relatively many GPUs). To efficiently utilize such performance, applications must scale up and efficiently utilize the heterogeneous computing architectures.

Many inefficiencies on large HPC systems are based on the lack of efficiency in the movement of data. On exascale systems, the size of the data involved necessarily increases, whether for simulations or analyses. This brings efficiency problems in the migration of data through the memory hierarchies, the offloading of data to GPUs, the communication of data between components with distributed memory and the storage of data in file systems. The aim of the DaREXA- F project is to mitigate these problems in simulations, which are of central importance for the development of fusion power plants and can significantly accelerate them, through data reduction measures and thus significantly increase the efficiency of the application of the GENE code developed by us and used worldwide. Existing systems, such as SuperMUC-NG, as well as future architectures in emulation and modeling are to be covered.

GENE is used to calculate complex physical processes in extremely hot fusion plasmas. By increasing the efficiency and scalability of GENE, it will be possible for the first time to calculate turbulent flows (and the associated heat and particle transport) within the entire plasma volume for large experiments (such as ITER). This would be a breakthrough in plasma physics with great visibility worldwide and many completely new application options, which will ultimately lead to a noticeable acceleration of fusion research.