This paper focuses on the implementation of vertex transformation and triangle clipping in the field of 3D acceleration. A dedicated matrix accumulating architecture is proposed to multiply transformation matrices and generate an accumulated transformation matrix. Then the accumulated transformation matrix is used for per vertex transformation. A combined clipping algorithm of Cohen-Sutherland and Sutherland-Hodgeman is adopted to accelerate triangle clipping. In addition, the clipping process is pipelined in order to increase the throughput. Experimental results show that our implementation can reach more than 25% performance higher than traditional method.