Energy use and carbon emissions in high-performance computing: A case study for universities and reduction strategies

Abstract

Energy use and greenhouse gas (GHG) emissions from high-performance computing (HPC) and high-volume data storage (HDS) strongly increased over recent years. This contrasts with the need for emission reductions to halt global warming in line with the Paris climate goals, calling for ambitious action also for these energy-intensive services. Here we focus on options for institutional users that aim at professional institutional climate action management, where the quantification of emissions from HPC and HDS as part of emission inventories is still rare. We present methods and results of a case study, drawing from HPC/HDS usage data of a collaborative consortium project of data centers and Universities in Austria, where institution-level quantifications of energy use and related GHG emissions were obtained. We find emissions from HPC use strongly dominating, relative to HDS use, for individual institutions and in total; case-study operational emissions summed to about 300–500 tCO<inf>2</inf>eq annually from HPC while HDS emissions were near 3 tCO<inf>2</inf>eq; co-estimated embodied emissions sum up to 50 to 100 and 3 to 10 tCO<inf>2</inf>eq, respectively. Building on the results, the collaboration of data centers and users enabled to derive also more general climate action options, including provider-user synergies. Provider improvements on energy savings, efficiency, shift to renewable sources and transparency to users on energy sources and consumption can meet with deliberate user choice of genuinely “green” data centers and user skill advances in “green coding”, “smart avoidance” of inefficient computations and considerate HDS management. Our findings underpin that providers and users jointly need to bring energy use and emissions under control to help meet the Paris climate goals.