Actually the throughput of integer instructions isn't that much lower than floating point operations, at least for arithmetic and bitwise operations.

http://docs.nvidia.com/cuda/cuda-c-programming-guide/#maximi...