Vector type library subroutines often perform much more efficiently than scalar type routines. The overhead expenses of scalar type routines are often comparable with the total time required for vector type computations, especially for highly optimized RNGs. To reduce overhead expenses, all VS RNG subroutines are of vector type.
Although vector type RNGs may require more careful programming, their use results in a substantial speedup in the overall application performance. VS provides a number of services to make vector programming as easy as possible. For further discussion, see the
Independent Streams. Leapfrogging and Block-Splitting
section of this document.
Vector type interface is useful for Monte Carlo methods because Monte Carlo requires a lot of random numbers rather than just one. You can call a vector RNG subroutine to generate a single random number as well, but such use is much less efficient.