Geocoding is crucial for using remotely sensed data in almost all applications, especially when a combination of multiple data sources is required. However, geocoding a synthetic-aperture radar image with the standard range-Doppler method is a time-consuming process due to unavoidable iterations. Using a replacement sensor model, for example the rational polynomial camera (RPC) model, can significantly reduce the calculation time cost. Another way to improve the calculation efficiency is to use massively parallel processing techniques. Modern graphics processing units (GPU) can be used as parallel processing units for computationally intensive applications. Using NVIDIA's Compute Unified Device Architecture (CUDA) the implementation of the range-Doppler and RPC methods on GPUs is easy, because the existing C/C++ code can be reused. With further optimizations for GPU processing, tremendous improvements can be achieved. The CUDA implementations run about 10 to 30 times faster compared with similar implementations on the central processing unit (CPU), and almost 200 times faster if particularly optimized for GPU computing.