This talk will present the case for a global large eddy simulation

model applying the super-parameterization (SP) methodology on

massively parallel computers. I proposed SP about 15 years ago to

improve representation of deep convection and accompanying cloud

processes in large-scale models of weather and climate. The main

idea behind SP is to embed---in all columns of the large-scale model

with a horizontal gridlength of the order of 100 km---copies of a

two-dimensional convection-permitting small-scale model with about

1 km horizontal gridlength and periodic lateral boundary conditions,

and to couple them to the outer model. This methodology can be

expanded by applying a high-spatial-resolution three-dimensional

SP model, essentially a large-eddy simulation model, and by embedding its copies in all columns of a large-scale model with the horizontal gridlength in the range of 10 to 50~km.  The outer model will then simulate processes down to the mesoscale (e.g., organized convection) and small-scale processes (e.g., boundary layer turbulence, convective drafts) will be simulated by LES models. Although significantly more expensive than the traditional SP, the SP LES is ideally suited to take advantage of parallel computers (e.g., applying GPU technology) because of the minimal communication between LES models when each processor runs a single LES model. Additional benefits of such a methodology will be discussed, and a simple computational example will be presented.