Power Consumption of Mixed Precision in the Iterative Solution of Sparse Linear Systems

This paper presents a detailed analysis of a mixed precision iterative refinement solver applied to a linear system obtained from the 2D discretization of a fluid flow problem. The total execution time and energy need of different soft- and hardware implementations are measured and compared with those of a plain GMRES-based solver in double precision. The time and energy consumption of individual parts of the algorithm are monitored as well, enabling a deeper insight and the possibility of optimizing the energy consumption of the code on a general-purpose multi-core architecture and systems accelerated by a graphics processor.