Efficient Acceleration for Large Scale Applications on HPC Heterogeneous Platforms by Carlos Hernandez (Univ. I Santander, Colombie)

HPC heterogeneous architectures with manycores, coprocessors and multi-accelerators consume high power and some applications underuse this computation ressources. To define how we can accelerate these applications is necessary to capture the consumption and to develop strategies for efficient acceleration and execution. This talk presents two monitors to capture energy/computing measures for NVIDIA GPUs and Intel Xeon Phi MICs called EnerGyPU and enerGyPhi. On the other hand, using the datalog of executions with enerGyPU and enerGyPhi monitors we propose strategies for efficient acceleration considering the workload of large scale applications executed in HPC Heterogeneous Platforms based on NVIDIA TESLA and Intel Xeon Phi Accelerators.