Abstract: To render a recursively defined subdivision surface by exploring the parallel computing power of GPU in computer graphics,an efficient algorithm for rendering the true limit of a face-split subdivision surface on GPU is presented in this paper.A reusable subdivision lookup table is generated off-line,which is essentially composed of the subdivision matrices.The size of the table is linear to the valences of vertices and the maximum subdivision depth,rather than exponential to them,and it is independent of the input mesh.In the runtime,the control patch containing two or more than two irregular vertices needs one-level pre-refinement.Then,the irregular patches are refined in parallel via the subdivision lookup table.In this way,the control points of the patches in each subdivision depth can be evaluated non-recursively,and parallelism of GPU is fully explored.Finally,all regular patches generated are rendered directly with hardware tessellation shader.The experiments show that the algorithm can render the true limit of subdivision surfaces efficiently.