two ideas we could try: - Use larger bitonic sorting networks (256 and 512 elements) - Improve pivot selection by pich a set of random indices