An efficient and robust pupil tracking system is an important tool in visual optics and ophthalmology. It is also central to techniques for gaze tracking, of use in psychological and medical research, marketing, human–computer interaction, virtual reality and other areas. A typical setup for pupil tracking includes a camera linked to infrared LED illumination. In this work, we evaluate and parallelize several pupil tracking algorithms with the aim of accurately estimating the pupil position and size in both eyes simultaneously, to be applied in a high-speed binocular pupil tracking system. To achieve high processing speed, the original non-parallel algorithms have been parallelized by using CUDA and OpenMP. Modern graphics processors are designed to process images at high temporal frequencies and spatial resolution, and CUDA enables them to be used for general-purpose computing. Our implementation allows for efficient binocular pupil tracking at high speeds for high-resolution images (up to 988 fps with images of 1280 × 1024 pixels) using a state-of-the-art GPU.