Finally we compare the render speed of different shaders. All timings have been done for a screen-filling mesh with 20k triangles (results for other triangle counts are similar).
We would like to point out some particular results from Table 12(a). For simple Phong shading without bump mapping or texturing, the shader with register combiner normalization runs at 80% of the speed of standard Gouraud shading, although much better visual quality is obtained (Figure 9). The shader with 16 bit cube map normalization and texture shader dot product which delivers the best image quality (Figure 10) runs with about half the speed of Gouraud shading. If shading with a decal texture and bump mapping is enabled, the shader with register combiner normalization runs also with slightly less than half the speed of standard Gouraud shading although the visual detail and shading quality are much better (Figure 11).