Cubed-Sphere Finite-Volume Dynamical Core (fvcore)
As horizontal resolutions of global climate and numerical weather prediction
applications increase beyond the hydrostatic limit of 10-km, the demand
for efficient scalable applications becomes essential. Over the next
decade global models will require scalability beyond hundreds of thousands
to millions of processors. The current implementation of the finite-volume
dynamical core (fvcore), a key component of global models at NASA (GEOS),
NOAA (AM) and NCAR (CAM/CSM), is limited in terms of scalability by the
choice of computational grid (latitude-longitude based). To break through
these scalability limitations the fvcore has been implemented on the
quasi-uniform cubed-sphere grid with a 2-dimensional horizontal domain
decomposition.
The cubed-sphere fvcore scales well beyond the pure MPI limitations
of the lat-lon core displaying ideal scaling on a full node
of the SGI Altix (512 CPUs). This improved scalability will
allow parent models to scale in excess of 10,000s of processors
as we move toward petascale computing environments.
+ Read More
+ Scientific Results
+ Computational Results