Platform for Scientific Computing

Platform for Scientific Computing at Bonn-Rhein-Sieg University


Platform for Scientific Computing

The Platform for Scientific Computing (Plattform für Wissenschaftliches Rechnen) is an interdisciplinary effort of professors from several disciplines and departments at the Bonn-Rhein-Sieg University to foster research and application on High Performance Scientific Computing (HPC). Main application areas are: Participating professors are:


The existing HPC hardware has several components each suitable for different computational demands. The main hardware ressources are, with a total of approx. 5,000 hardware threads, approx. 200,000 GPU-cores, an accumulated main memory size of approx. 17.1 TB, and an accumulated disk space of approx. 785 TB (gross): Nodes are connected with 200 Gb/s Infiniband, 100 Gb/s Omni-Path and/or 1/10Gb-Ethernet. A detailed description of the hardware configuration can be found here. Some pictures can be found here.


The front-end node is the only external visible access point and can be accessed through The nodes can be used with parallel programs that are either based on a shared memory programming model (e.g. OpenMP) using up to 256-way parallelism, based on a GPU programming model (CUDA, OpenCL, OpenACC) using up to approx. 30,000 GPU cores in a system, or the cluster computer can be used in a distributed memory model (e.g. MPI) using all cluster nodes with up to approx. 5,000-way parallelism.

A detailled descriptions on how to use the cluster and information on the software configuration can be found here .

Current Status

The current system status can be seen here. Pages get updated every minute.

Access and Contact

Access is granted for research and educational projects supervised by any of the participating professors. Additionally, other projects are possible and explicitly encouraged. Please feel free to ask questions. Contact person is Prof. Dr. Rudolf Berrendorf .


The Platform for Scientific Computing is/was supported by research grants: