Performance changes between 1.2.1 and 1.3.1

Hello, I have updated the version from 1.2.1 to 1.3.1. The training process keeps all cores busy but does not seem to be able to finish, or it has become much slower. The list of changes in 1.3.0 is quite long, and I am currently looking for ideas regarding what the problem could be and suggestions about what I could try to adjust to narrow it down. Thank you!

https://github.com/dmlc/xgboost/issues/6552 might be related.

Thank you! That must be it. I am glad someone has managed to put together an example.