Traditionally a rotary forging process is a kind of metal forming method where a conic upper die, whose axis is deviated an angle from the axis of machine, forges a billet continuously and partially to finish the whole deformation. For the rotary forging process simulation, more re- searches were focused on simulating the simple stage forming process with axisymmetric part geometry. Whereas in this paper, the upper die is not cone-shaped, and the billet is non-ax- isymmetric. So the movement of the punch is much more complicated than ever. The 3D FEM simulation models for the preforming & final forming processes are set up after carefully study- ing the complicated movement pattern. Deform-3D is used to simulate the material flow, and the boundary nodal resisting forces calculated by the final stage process simulation is used to ana- lyze the final forming die strength. The CAE analysis of the die shows that the design of the fi- nal forming die is not reasonable with lower pre-stress which is easy to crack at the critical comers. An optimum die design is also provided with higher pre-stress, and verified by CAE analysis. Traditionally a rotary forging process is a kind of metal forming method where a conic upper die, whose axis is deviated an angle from the axis of machine, forges a billet continuously and partially to finish the whole deformation. More re- searches were focused on simulating the simple stage forming process with axisymmetric part geometry. Whereas in this paper, the upper die is not cone-shaped, and the billet is non-ax- isymmetric. So the movement of the punch is much more complicated than ever. The 3D FEM simulation models for the preforming & final forming processes are set up after careful study- ing the complicated movement pattern. Deform-3D is used to simulate the material flow, and the boundary nodal resisting forces calculated by the final stage process simulation is used to ana- lyze the final forming die strength. The CAE analysis of the die shows that the design of the fi- nal forming die is not reasonable with lower pre-stress which is easy to crack at the critical comers. An optimum die design is also provided with higher pre-stress, and verified by CAE analysis.