This is a GPU implementation of fractional differencing (we call it GFD). It allows rapid large-scale implementation of fractional differencing to minimize memory loss while achieving stationary for time series data.