The presented new rendering algorithm for maximum-intensity projection (MIP) runs at real-time frame rates. In combination with the preprocessing step which removes parts of the volume which do not contribute to a MIP image and the sorting of the remaining voxels by value within the RenderList storage scheme, overhead for traversing empty regions is eliminated. Compared to other optimized MIP algorithms, up to an order of magnitude is gained in speed, compared to a brute-force ray casting approach even two orders of magnitude. A preview mode for interactions on low-end hardware is provided basically ``for free'' by skipping voxels with low data values. Two extensions of the algorithm allow to generate depth-shaded MIP and LMIP at comparable speed.