Accelerated MR physics simulations on multi-GPU systems

A multi-GPU approach of MRISIMUL, a recently developed step-by-step comprehensive MR physics simulator of the Bloch equation, is presented in this study. The specific aim was to apply MRISIMUL on multi-GPU systems so as to achieve even shorter execution times. We hypothesized that such a simulation platform could achieve a scalable performance with the increasing number of available GPU cards on single node, multi-GPU computer systems. A parallelization strategy was employed using the MATLAB single-program-multiple-data (spmd) statement and an almost linear speedup was observed with the increasing number of available GPU cards on two separate systems: a single computer of 2 quad-core processors and two Tesla C2070 GPU cards and a single computer of 2 hexa-core processors and four Tesla C2075 GPU cards.