An important computational cost is associated with General Circulation Model (GCM) simulations. When several simulations of the same model must be run and can be parallelized, it is desirable to distribute the load over several HPC platforms in order to decrease the effective time of simulation. The members of CMIP5 historical simulations of the EC-Earth GCM were run, for example, in different centers. However, distributing the total load on different platforms can only be done if differences in hardware do not introduce differences in the results. The Earth Sciences Department at Barcelona Supercomputing Center (BSC-CNS) is currently investigating the validity of this hypothesis.


Slides to this talk