Main Content

High performance computing

interne Computerchips
Photo model: Colourbox.de

Due to the constantly growing requirements in the field of Big Data and AI, a powerful HPC infrastructure is needed. For this reason, the Marburg Computing Cluster (MaRC3) was established. The computing cluster was built in cooperation with different research groups of the Philipps-Universität Marburg in order to use the hardware resources cooperatively and efficiently.
In this context, the Department of Mathematics and Computer Science contributed a large number of high-performance graphics processing units (GPUs) in order to be able to meet the extremely increased demands of both machine learning and deep learning as well as classical HPC tasks. Especially in the area of Deep Learning, developments are moving towards ever deeper and more computationally intensive network architectures as well as ever larger data sets. By using the GPUs integrated in MaRC3, the training times of current network architectures can be reduced from several weeks to a few days or hours, and the inference or analysis of large data sets can also be drastically accelerated.

Currently, the department has 68 Nvidia graphics processors for data centers:

  1. 16 NVIDIA A100 Tensor Core GPUs with 40 GB GPU memory
  2. 52 NVIDIA A100 Tensor Core GPUs with 80 GB GPU memory

Access to the hardware resources is controlled via the so-called SLURM Workload Manager (Simple Linux Utility for Resource Management).
Here, the scientists and researchers of the Department of Mathematics and Computer Science have prioritized access to the GPU hardware resources of the department via so-called "owner queues".
Further information about MaRC3.