The Global/Regional Assimilation and PrEdiction System (GRAPES) is a newly developed global nonhydrostatic numerical prediction model, which will become the next generation medium-range operational model at China Meteorological Administration (CMA). The dynamic framework of GRAPES is featuring with fully compressible equations, nonhydrostatic or hydrostatic optionally, two-level time semi-Lagrangian and semi-implicit time integration, Charney-Phillips vertical staggering, and complex three-dimensional pre-conditioned Helmholtz solver, etc. Concerning the singularity of horizontal momentum equations at the poles, the polar discretization schemes are described, which include adoption of Arakawa C horizontal grid with v at poles, incorporation of polar filtering to maintain the computational stability, the correction to Helmholtz equation near the poles, as well as the treatment of semi-Lagrangian interpolation to improve the departure point accuracy, etc. The balanced flow tests validate the rationality of the treatment of semi-Lagrangian departure point calculation and the polar discretization during long time integration. Held and Suarez tests show that the conservation properties of GRAPES model are quite good.