—This paper proposes a programmable geometry engine (GE) reducing the expensive internal buffers and register files of the conventional programmable GEs and sharing datapaths of a special function unit. The proposed GE is appropriate for the embedded 3D graphics environment where the reduction of hardware cost is a critical issue. The degraded performance caused by the hardware reduction is compensated by a variable write-back timing architecture with a dynamic hazard controller and a data forwarding method. The GE is implemented by a 0.13um CMOS technology and has the performance up to 33.2Mvertices/sec,