New Projects for Energy-Efficient HPC

JSC is participating in the BMBF funding track “Energy-Efficient HPC (Green HPC)” with two research projects. Data centres operating in Germany currently consume around 3 % of the total electricity produced and are responsible for CO₂ emissions of just over 8 million tonnes per year. As a result, even small energy savings ultimately lead to significant reductions in the total energy consumption of data centres and thus to relevant savings in CO₂. The aim of this funding track is to improve the energy efficiency of high-performance computing, both of high-performance computing centres at universities and research institutions and of commercial computing centres. Both of the projects involving JSC are funded for three years as of 1 October 2022.

The ENSIMA project (ENergy-efficient SImulation Methods for Application-oriented computational problems) will use AI methods to improve the determination of design parameters and accelerate the execution time of simulation processes through approximate and heterogeneous computing. For the sample use case – sheet metal forming in the automotive industry – the goal is to reduce computing time by 50 %, leading to a 15 % reduction in steel usage and thus indirectly to a reduction in both manufacturing-related emissions and energy requirements for vehicle production. FZJ will lead work package 2 (“Optimization of Workflow Components”), which will analyse and improve the performance and energy efficiency of the OpenForm user code components used in the project, which is supplied by industry partner GNS. Further project partners are RWTH Aachen University (coordinator) and TU Darmstadt.
Contact: Dr. Bernd Mohr

The STXDemo project will provide a proof of concept for future energy-efficient high-performance comput­ing. It will develop a hardware prototype made of a motherboard with multiple STX processors, which are processing units optimized to solve stencil algorithms. STXDemo will build on the work done in the EU European Processor Initiative (EPI) project and aims to improve energy efficiency by at least a factor of two compared to competing platforms based on GPUs or x86 processors. The benchmark for the success of the project is the improvement of the energy efficiency of selected computationally intensive simulation applica­tions. FZJ will contribute a co-design application from the field of lattice quantum chromodynamics (LQCD) and will host one hardware prototype at JSC. Further project partners are the Fraunhofer-Gesellschaft (coordinator), TU Kaiserslautern, and Zollner Elektro­nik AG.
Contact: Dr. Estela Suarez

from JSC News No. 293, 6 December 2022