I think may be it depends on the code which you have written, could you please your example . Spark dataframe under the hood calls sql only .. so in most of the scenarios it should be same