Expanding the boundaries of GPU computing