It's possible that FAH could set up separate MPI ranks for E cores and P cores, but by default it only uses a single rank, and the load balancer can't do much to balance loads within a rank, and the performance difference between P cores and E cores is so vast that the load balancer has no chance. It's much better at balancing loads between ranks. I could give this a test and see how well GROMACS performs on a hybrid system using separate MPI ranks and suggest that internally, but there are a few reasons why I wouldn't expect fast results:
- Detection of hybrid systems is fragile and the code to do it reliably isn't in any of FAH's libraries yet
- FAH has extensively tested single-rank folding and has little experience with multi-rank folding
- Developer time is stretched very thin and not many people are working on the GROMACS FAH cores