High Performance Computing (HPC) systems are a type of research infrastructure that drive development across a wide range of research topics within science and engineering. The drive for ever more complex models or larger scale evaluations creates a constantly growing demand for computing performance. For technical and financial reasons, there is an international consensus that the power dissipation of future HPC systems will be practically limited at 20 MW, which is just above the power dissipation of current leading systems. Thus, future performance improvements must come through improving the energy efficiency of HPC infrastructures and applications. Optimizing HPC applications for energy efficiency is difficult for two reasons. First, how to best design a validated, accurate and practically realizable energy measurement infrastructure for large-scale HPC systems is still an open research problem. Second, the engineering overhead of optimizing for energy efficiency is too high.
To alleviate these challenges, NTNU has decided to build the “Energy efficient high Performance computing research InfrastruCture (EPIC)”. EPIC will enable research that can identify energy measurement strategies with acceptable accuracy that are implementable in a large-scale cluster. In EPIC, we will also develop software that simplifies the task of optimizing HPC applications for energy efficiency. Finally, we will use EPIC to provide novel, energy efficient implementations of important HPC applications.