I've seen it done on a few client sites. We "borrow" the gpu machine and run Fluent on the cpu & Rocky on the gpu: it's more efficient and avoids conflicts with our queue system.
One issue is queuing software. There's an option in Rocky coupling to tell Fluent where to run. So, most issues have been with cluster config rather than coupling. I'll see if one of the Rocky team can comment. Error messages will help, along with the HPC OS and whether you're using schedulers etc.
You are navigating away from the AIS Discovery experience