We propose a dynamic program coupled with finite elements for valuing two-dimensional American-style options. To speed-up our procedure, we use parallel computing at every step of the recursion. Our model is flexible because it accommodates a large family of option contracts signed on two underlying assets that move according to a lognormal vector process. The same procedure can be adapted to accommodate a larger family of derivative contracts and state-process dynamics. Our numerical experiments show convergence and efficiency, positioning our method as a viable alternative to traditional methodologies based on trees, finite differences, and Monte Carlo simulation.