Rapid large-scale fractional differencing with RAPIDS to minimize memory loss while making a time series stationary. 6x-400x speed up over CPU implementation.