Kick-off workshop to prepare for a new AI cluster at MPCDF

Eleven Max-Planck Institutes, together with the MPCDF have joined forces to deploy a dedicated GPU machine specifically for AI applications. A corresponding kick-off workshop was recently hosted by the MPCDF.

How AI is changing our research – and what we need to use It effectively? A new, jointly operated AI cluster is set to take research within the Max Planck Society to the next level. A recent workshop at the MPCDF laid the foundation for technology, collaboration, and knowledge transfer. Artificial intelligence is playing an increasingly important role in the research conducted at the Max Planck Society. Whether it’s training neural networks on specific datasets or fine-tuning large language models, the demand for computing power is growing rapidly. Even the "simple" task of running modern AI models (inference) now requires specialized hardware with substantial GPU memory. To meet these demands, eleven Max Planck Institutes have come together, initiated by the MPCDF. With support from the BAR, they are jointly procuring a dedicated AI cluster, which is scheduled to go into operation at the MPCDF starting in May/June 2025.

Beginning of April, representatives from the participating institutes met in Garching to prepare for the deployment of the new cluster. The focus was on two key questions:

• What requirements do the institutes have for the system? In short presentations and discussion sessions, various usage scenarios were presented – ranging from training to inference of large models. The goal: a balanced usage concept that serves everyone.
• How can knowledge transfer be improved? Many institutes are working on similar AI-related topics – such as the fine-tuning of large language models. The idea of an “AI Knowledge Hub” was introduced: a platform for sharing, training, and enhanced collaboration. Initial communication channels and software solutions were discussed, as well as the desire for AI training courses. Suggestions on this will soon be provided by the MPCDF.

Participants included representatives from seven participating Max-Planck Institutes (MPI) – the MPI for Informatics, MPI for Software Systems, MPI for Sustainable Materials, MPI for Biogeochemistry, MPI for Human Development, MPI for Human Cognitive and Brain Sciences, and the Fritz Haber Institute of the Max Planck Society – as well as interested guests from the MPI of Quantum Optics and the MPI for Neurobiology of Behavior – caesar.

Contact: Erwin Laure, Andreas Marek