Power-aware Performance of Mixed Precision Linear Solvers for FPGAs and GPGPUs