Calculating variance of Quasi-Monte Carlo with scrambled Sobol

I am trying to use the host API of cuRAND to generate a scrambled Sobol sequence for use in a Quasi-Monte Carlo simulation for pricing and calculating Greeks of a few different option types.

It is my understanding that if I want to calculate the overall variance, I need to calculate the estimates from each scrambled sequence separately and then compute the variance from the averages of those estimates.

Currently, I am generating NPATHS * NTIMESTEPS random numbers with the CURAND_RNG_QUASI_SCRAMBLED_SOBOL32 generator with the number of dimensions as NTIMESTEPS. How do I know which random numbers have been generated from which scrambling set?

If this is not possible from the host API, how would I map the scrambling constants to the random numbers generated if I were to use the device API?