Overcoming the GPU memory limitation on FDTD through the use of overlapping subgrids

The method Finite Difference Time Domain (FDTD) is widely used in electromagnetic simulations. Since this method is a data intensive and computation intensive problem, there are a lot of initiatives to improve the scalability and the performance of the FDTD. Specifically the use of GPU to accelerate the FDTD is in focus, which has a good cost-benefit, offering a speedup of hundreds of times if compared to the traditional CPU computation. Nevertheless, the current implementations of FDTD in GPU need GPU devices with memory capacity to allocate the roll data needed for the simulation, which in many cases could require additional investments to increase the domain of the simulation. This work proposes a solution of FDTD over GPU that uses overlapping subgrids to increase the domain of the FDTD simulation through the use of CPU memory. The redundancy of the overlapping subgrids is used to minimize the memory traffic between GPU and CPU, increasing the performance. Tests of performance show that the solution combines the power of GPU with the memory facility of the CPU, allowing the accomplishment of simulations that require more memory than what is offered by the GPU device at a damaging performance cost lower than 10%.