During the user-group meeting it was suggested that providing benchmarking (and example job submission scripts) for single-GPU, multi-GPU and multi-node multi-GPU.
HECBioSim have some benchmarking for AMBER/GROMACS etc. on their website, for single-node single/mulit-gpu benchmarks, which could be linked references on the amber/gromacs page(s).
At least if you run a series of cases, multi-node (even multi-GPU) gromacs/amber normally isn't useful if the problem doesn't require it, because of scheduling throughput, even with perfect scaling.
During the user-group meeting it was suggested that providing benchmarking (and example job submission scripts) for single-GPU, multi-GPU and multi-node multi-GPU.
HECBioSim have some benchmarking for AMBER/GROMACS etc. on their website, for single-node single/mulit-gpu benchmarks, which could be linked references on the amber/gromacs page(s).