On the field of parallel programing has emerged a new big player in the last 10 years. The GPU's have taken a relevant importance on scientific computing because they offer a high performance computing, low cost and simplicity of implementation. However, one of the most important challenges is the program languages used for this devices. The effort for recoding algorithms designed for CPUs is a critical problem. In this paper we review three of principal frameworks for programming CUDA devices compared with the new directives introduced on the OpenMP 4 standard resolving the Jacobi iterative method.