The calculation time is greatly depend on signal processing unit I think.
The 1000-point FFT is not the most optimal decision I think. It would be better to make a data vector a little bit longer - to the next power of 2, i.e. 1024.
For power of 2 values FFT complexity is minimal.
Instead of calculation time it wold be better to operate such a value as number of operations.
And at last you are free to use google to find out what are these operations and how many of them it takes.