I'm surprised to have not seen a mention of the python library Dask in this thread. Whilst I agree that reading CSVs in Python could be a lot quicker, most people do tend to use third party libraries to do so. Dask is able to multi-core multi-thread almost all operations, including reading in a CSV. You can go from a Dask data frame to a pandas dataframe.