there are matrix multiplication routines developed for CUDA. OpenCL you would have to do everything from the ground up. So Nvidia gave everyone a head start for numerical computation, and that edge has ever since snowballed.