Image-based CSG rendering algorithms for standard graphics hardware rely on multipass rendering that includes reading and writing large amounts of pixel data from and to the frame buffer. Since the performance of this data path has hardly improved over the last years, we describe new implementation techniques that efficiently use modern graphics hardware. 1) The render-to-texture ability is used to temporarily store shape visibility, avoiding the expensive copy of z-buffer content to external memory. Shape visibility is encoded discretely instead of using depth values. Hence, the technique is also not susceptible to artifacts in contrast to previously described methods. 2) We present an image-based technique for calculating the depth complexity of a CSG shape that avoids reading and analyzing pixel data from the frame buffer. Both techniques optimize various CSG rendering algorithms, namely the Goldfeather and the layered Goldfeather algorithm, and the Sequenced-ConvexSubtraction (SCS...