Before anyone gives us a huffy, "That was so last year" Alicia Silverstone impression, let us say that per pixel shading is the complement to the vertex shader NVIDIA wishes could have been included with the GeForce2 GTS. Unfortunately, the complexity of the vertex shader kept it from being developed until now, and even still, the GPU weighs in at 57 million transistors!

After the custom effects have been applied to a scene via the vertex shader, the pixels that comprise the scene still need to be rendered, lit, shaded and colored. Previously, low-resolution texture maps would be applied to polygons to create a game's scenery. With the GeForce2 GTS, developers were given the ability to control lighting effects on a per-pixel level through the NVIDIA Shading Rasterizer with fixed-function pixel pipelines.

With a programmable pixel shader, the GeForce3 allows developers to create their own pixel-shading effects rather than adhere to the more limited options provided by the NSR. Whereas the GeForce2 could only process two texture-blend operations per pass, the GeForce3 can deliver up to 8 of these blending operations per pass (four pipelines with two texture units each). Moreover, a maximum of two pixels can have four texture applied per pass, unlike the GeForce2 GTS that required multiple passes to apply the additional textures. This helps alleviate bottlenecking problems in the chip since multiple passes require extra geometric transformations and Z-buffer calculations.

Further expanding the pixel shading capabilities of the GeForce2, NVIDIA has enabled up to 36 operations running in parallel, as opposed to just seven on the GeForce2 GTS. Hardware shadows, reflective bump mapping, material properties, and per pixel lighting all benefit from this expansion. Some of the effects that NVIDIA demonstrated to us were skin, hair, fur, and realistic reflections.