Solving the cardiac bidomain equations using graphics processing units

► We present a GPU implementation of the cardiac bidomain equations based on sophisticated methods such as CG and multigrid. ► A single GPU accelerated the cardiac simulator by 7 times compared to the best performance obtained using a single CPU. ► A single desktop computer equipped with a GPU is 1.4× faster than a parallel implementation running on a cluster of 16 cores. ► Numerical errors associated with the single-precision operations on GPU may impair the solution of the bidomain equations. ► A detailed performance analysis points out the tasks that deserve better GPU implementations on future works.