Is std::mt19937_64 faster than std::mt19937?

Does mt19937_64 have a higher throughput (bit/s) than the 32 bit version, mt19937, assuming a 64 bit architecture?

What about after vectorization?