It is worth noting that the database setup for gpu mode is slightly different than cpu and an automatic setup subworkflow is not in the pipeline yet so it is assuming that your dbs are compatible with the gpu implementation.