It would seem to me that this kind of program would be heavily reliant on branching for which Vector processing (which is what GPUs excel at) is not the best choice.