Large-scale biophysically detailed computer models of the heart provide a useful tool to understand dynamics of cardiac excitation and mechanisms underlying lethal cardiac arrhythmias. However, high demanding of intensive high performance computing environments limits the practical application of such models. This paper presents a novel use of a desktop personal computer and the CUDA parallel computing architecture for a highly efficient method of parallel simulation of a 3D ventricular model. We show that substantial speed increases can be obtained using a desktop Graphical Processing Unit (GPU) compared to a single desktop Central Processing Unit (CPU), and that a single GPU can be an effective substitute to large numbers of CPUs.