Evaluation of low-power architectures in a scientific computing environment

HPC (High Performance Computing) represents, together with theory and experiments, the third pillar of science. Through HPC, scientists can simulate phenomena otherwise impossible to study. The need of performing larger and more accurate simulations requires to HPC to improve every day. HPC is constantly looking for new computational platforms that can improve cost and power efficiency. The Mont-Blanc project is a EU funded research project that targets to study new hardware and software solutions that can improve efficiency of HPC systems. The vision of the project is to leverage the fast growing market of mobile devices to develop the next generation supercomputers. In this work we contribute to the objectives of the Mont-Blanc project by evaluating performance of production scientific applications on innovative low power architectures. In order to do so, we describe our experiences porting and evaluating sate of the art scientific applications on the Mont-Blanc prototype, the first HPC system built with commodity low power embedded technology. We then extend our study to compare off-the-shelves ARMv8 platforms. We finally discuss the most impacting issues encountered during the development of the Mont-Blanc prototype system