Combining multiple CPUs and single GPU
 

Generally speaking, are there CFD code development strategies for utilising multiple CPUs and a single GPU, making it more efficient/faster than utilising just the multiple CPUs or just the single GPU (with single CPU)?

An example strategy would be: let's say I partition a mesh across X CPU cores. Each CPU core then utilises 1 of X thread blocks (using CUDA) to solve equations on respective partitioned mesh section.

Instead of doing this, would it generally be better to either just use the multiple CPUs alone or just the single CPU/GPU?