The GPU driven journey towards more sustainable HEP computing

Energy efficient usage of our hardware resources has become a topic of ever-increasing importance. Both, high operational costs and environmental concerns are good reasons for us to optimize the way we use hardware. In comparison with CPUs, GPUs are a convenient alternative that has the possibility to achieve a significantly higher energy efficiency in many of the HEP relevant workflows, like simulation, reconstruction, or machine learning based end user analyses. This talk aims to shed some light on how efficient such applications run on different hardware, using the example of a number of GPU benchmarks that have been performed on the Karlsruhe Tier 3 cluster TOpAS.