HPC Cluster Sofja: High Performance Computing and Data Science
In the field of computational modeling and data-driven quantitative analysis, the application of machine learning methods, in particular "deep learning", is central. These methods allow solving highly complex pattern recognition and predictive information processing tasks through the data-driven optimization of deep neural networks and related nonlinear systems. Therefore, computational research in the field of biological network dynamics requires a powerful HPC platform.
All members of the CIDBN can use those HPC resources of the GWDG that are generally accessible for their computationally intensive simulations and data analyses. In order to be able to meet the foreseeable demand for computing resources for research at CIDBN, CIDBN has procured a new dedicated HPC cluster that is hosted by GWDG.
- The servers in this HPC cluster are all equipped with AMD CPUs.
- 30 computing nodes with two AMD 7763 2.45 GHz processors each provide a total of 3840 cores.
- The scratch file system belonging to this cluster is about 500 TB in size, of which more than 100 TB are in the form of NVMe technology.
- Two interactive server nodes are used to access the cluster.
- All nodes in this cluster are interconnected by a high-speed Mellanox HDR 100 Infiniband network.
As part of the procurement of the system, we especially considered sustainability criteria in production and operation. The system is in operation since March 1st 2023. Read more about Sofja in in issue 4 of the GWDG News, pp. 13-15.
