In this instance neither df.display() or display(df) does work because in the end both will return the same sql to SQL Server. However I noticed it seems to be an error in the config of our cluster as it does work fine on a new compute. 

Probably has to do with the custom dialect we are using to map TimestampType to datetime2.