I will present work on porting the cubed-sphere dynamical core used by NASA/Goddard and NOAA/GFDL to the Intel Phi multi-core architecture. The dynamical core is a finite-volume method implemented on a cubed-sphere grid and represents a significant portion of the computational time used in simulations run at both organizations. We find that performance on the Phi requires: large-scale OpenMP concurrency, strong single-core performance utilizing vectorization and data alignment, and efficient MPI communication. Addition ally, for our work to result in a useful code-base for scientific production runs we require that we are able to maintain a single code-base for both the Intel Phi and conventional architecture that is maintainable and readable. Given these requirements and constraints, I will discuss our choices and work to satisfy these criteria. I will report our performance results and identify the limitations we've encountered.