NHR South-West is a collaboration of Goethe University Frankfurt, Johannes Gutenberg University Mainz, University of Kaiserslautern-Landau and Saarland University, embedded into the German National High Performance Computing (NHR) association.
Simulations and Data Labs
The different disciplines in life sciences have among the fastest growing demands concerning compute performance, data storage, and AI/ML support, while they also have a huge overlap with the physics discipline of molecular systems. An overarching topic of this SDL is multi-scale modeling, which requires new approaches to data organization and management. This SDL interacts closely with the SDL Biology at JSC and supports researchers from medicine. It focuses on the de- velopment and user support for scalable HPC methods and further supports machine learning for the various user groups and applications.
Simulations and Data Labs
The SDL combines the interplay between physical research from the domain of nuclear, particle, and astrophysics with the development and application of algorithms and software for current and future HPC architectures. The first topical focus will be on QCD applications, which require huge core counts, low latency interconnects and high-bandwidth memory. The computational scientists of this SDL will help to extend the established workshops series on „Lattice Practices“, which is currently organized by DESY/Zeuthen and Jülich, with an emphasis on the use of new hardware architectures to strengthen the community’s ability to make the most efficient use of the allocated computing resources. The second main focus of this SDL will be to assist users from astrophysics to optimize scientific codes such as Whisky or Llama to apply vectorization and GPU accelerators and to include new methods like the CCZ4 formulation.
Main focus of this Method Lab is to offer support in code optimization and to develop a tooling infrastructure for improving the efficiency of scientific software. Standard sup- port offered by this lab includes code optimization, vectorization, and adoption of the code for GPU use. Novel tools for this lab will also be made available by our other method labs.
AI is seeing a tremendous uptake across essentially all scientific disciplines and is growing exponentially in terms of novel AI techniques and available tools. The general goal for this method lab will be to maintain not only an overview about this development and help scientists navigate this space with respect to their specific needs, but to maximally support them in their use of AI/ML. We will specifically assist users with the visualization, debugging, and explainability of AI systems, which is becoming increasingly important to better understand and make efficient use of these technologies in concrete projects (a.k.a. Explainable AI).
Within this lab, we will adapt AnyDSL (Domain-specific Language) and related technology to the needs of users and offer general support. We will host regular tutorials and provide individual support to customers implementing DSLs in AnyDSL. Furthermore, we provide support in fundamental DSL techniques such as metaprogramming, DSL embedding, and the adoption of compilation frameworks and techniques such as LLVM. Additionally, we provide support for high-performance derivative computations for DSLs via automatic differentiation software such as CoDiPack.