Power-Aware Parallel 3-D Finite Element Mesh Refinement Performance Modeling and Analysis on CUDA/MPI Multi-core and GPU Architectures

摘要

The power dissipation of parallel 3-D mesh refinement is highly dependent on the underlying algorithm and the power-consuming features of the processing elements (PEs). This contribution presents a practical methodology for modeling and analyzing the power performance of parallel 3-D FEM mesh refinement on CUDA/MPI architectures based on detailed software prototypes and power parameters in order to predict the power functionality and runtime behavior of the algorithm, optimize the program design and thus achieve the best power efficiency. We propose approaches for GPU parallelization, dynamic CPU frequency scaling and dynamic load scheduling among PEs. The performance improvements of our designs are demonstrated and the results are validated on a real multi-core and GPU cluster.