Huge performance gap when processing ORC files created by Spark

Seeing a huge performance difference while running a simple filter query on ORC files created by Spark. I'm seeing better performance if the files are written by Hive i.e. after doing a "create table x as select * from y".