ProjectEnergy-efficient Computing

Energy and performance measurement of numerical algorithms and simulations

Simulations with high-performance computers (HPC) are an established pillar of scientific research along with theoretical analyses and experiments. At the same time, however, these simulations are highly energy-intensive, such that, after just a few years, the operating costs for the energy supply start to exceed the acquisition costs of a high-performance computer. Therefore, issues regarding energy and performance measurement of numerical algorithms and simulations in the field of high performance computing are of paramount importance to the project “Energy Efficient Computing.”

The goal is to define and perform benchmarks for the energy use and runtime of numerical algorithms and simulations. In order to do this, a measurement system consisting of energy measurement devices and a measurement server will be set up and configured to operate automatically. The measurements and their evaluation will be performed using suitable profiling tools.

In addition, numerical methods optimized for energy consumption will be developed. The first numerical algorithms will include parallel implementations of mixed-precision methods, geometric multigrid methods and asynchronous iteration methods for solving linear equation systems on graphics processing units (GPUs). Students have the opportunity to take part in this project through internships.