Analyzing the potential of GPGPUs for real-time explicit finite element analysis of soft tissue deformation using CUDA

As the presence of finite element implementations on General Purpose Graphics Processing Units (GPGPUs) is the literature increases, detailed and in-breadth testing of the hardware is somewhat lacking. We present an implementation and detailed analysis of an FE algorithm designed for real-time solution, particularly aimed at elasticity problems applied to soft tissue deformation. An efficient parallel implementation of Total Lagrangian Explicit Dynamics implementation is elucidated and the potential for real-time execution is examined. It is shown that in conjunction with modern computing architectures, solution times can be significantly reduced, depending on the solution strategy. The usability of the method is investigated by conducting a broad assay on ranging model sizes and different cards and comparing to an industry-proven FE code Abaqus. In doing so, we study the effect of using single/double precision computation, quantify and present error measurements as a function of the number of time-steps. We also examine the usage of a special texture memory space and its effect on computation for different devices. Adding material complexity in the form of a tissue damage model is presented and its computational impact elucidated. The aggregate results show that, for a particular set of problems, it is possible to compute a simple set of test cases 30-250 times faster than current commercial solutions. According to the speedups achieved, an indication is provided that the GPGPU technology shows promise in the undertaking of real-time FE computation.status: publishe