High Performance Computing (HPC) software is a critical component of scientific infrastructure, but it can only advance research if it is able to use the most advanced hardware effectively. Additionally, software must be constantly developed further in order to stay abreast of a changing hardware landscape. Individual users often lack the time and knowledge to improve their code, data centres lack staff, and automated improvements with tools are rarely successful.
Partners in the Enabling Performance Engineering for Hesse and Rhineland-Palatinate (EPE) project include the three Rhine-Main universities as well as the University of Kaiserslautern. The project will deepen the existing expert advice offered on the topic of HPC programming and code-tuning at the participating institutions. The existing infrastructures for knowledge distribution will be networked and expanded. The goal is to pass on the existing scientific HPC expertise to the users of the high-performance computers. This increases the performance of the application code and makes use of the high-performance computers more efficient. The focus is on model-based scaling analyses, GPU performance engineering, algorithmic stability and reproducibility of results in the HPC context.