Performance

In order to assess the performance of running on ARCHER2 ...

Scaling Tests

All standard output and hybrid MPI method unless otherwise stated

Bare-Metal

GNU/openMPI O0
GNU/openMPI O1
Cray/MPICH O0
Cray/MPICH O1
Cray/MPICH O1 No Output / Heavy Output

SIF

GNU/openMPI O0
GNU/openMPI O1
Cray/MPICH O0
Cray/MPICH O1

Addtional SIF

Cray/MPICH O1 Bloated
Cray/MPICH O1 No Output / Heavy Output
Cray/MPICH O1 Bind MPI method
Cray/MPICH O1 multi-stage + apps

Bare-metal

Graphs + optimisations

NEMO SIF

Graphs + optimisations

Summary

Thoughts

one
two
three