Comparison of GPU Architectures for Asynchronous Communication with Finite-Differencing Applications