It bears repeating that fillrate and subsequent comparisons were made between a 164 MHz Radeon and a 200 MHz GeForce2. Radeon contains two pixel pipelines. Each pixel pipeline receives inputs from three texturing units that are capable of compositing three textures in a single pass. GeForce2 contains four pixel pipelines. Each pixel pipeline receives inputs from two texturing units.

[It is worth mentioning here that the 3DMark 2000 benchmark program uses four textures for the fill rate tests. So if your hardware can do multiples of four textures per pass, you will see optimal results. Thus it does not make efficient use of Radeon's third texture unit. When future games and applications (and benchmarks) are making use of three textures, Radeon should show an advantage in this area. Until then, the common approach has been in twos. - ED]

When rendering in 16-bit color, fillrate measurements for Radeon suggest four texturing units to be active out of a possible six, with an output of two dual-textured pixels per clock. GeForce2, with six active texturing units (out of eight) and an output of three dual-textured pixels per clock, has a sizeable lead in 16-bit rendering (figure 1). If each pixel is composed with three textures instead of two, the contribution from all six of Radeon's texturing units should bring it on par with GeForce2.

In 32-bit renders, the architecture of GeForce2 only allows four texturing units to be active compared to six in 16-bit color. As for Radeon, the number of active texturing units remains the same, i.e. four. Therefore, GeForce2's fillrate drops dramatically while Radeon's fillrate drops marginally.