In the following, we'll take a closer look at the method of texture filtering that the Radeon X800 and the GeForce 6800 Ultra use. Both companies' cards use what's called an adaptive filtering technique, which means that the driver employs an optimized filtering method to offer higher performance. Therefore, the cards filter the textures differently according to their viewing angle in the plane. An extreme case of this would be where certain textures will only be filtered 2x despite the fact that the 8x filter is chosen by the user or application. Additionally, NVIDIA also uses an optimized trilinear filter that can, thankfully, now be switched off in the driver.

In practice, i.e. games, this method works very well, since the differences in the scenes are only visible upon very close scrutiny. In the end you get a way better performance with only little or no degradation of image quality, so it's completely okay for the average gamer.

Purists and enthusiasts will certainly see this in a different light - they want the maximum possible image quality for their maximum-priced acquisitions. Since ATi and NVIDIA also use adaptive techniques that differ in some details, this results in different performance when texture filtering is active. The attempt to make an "apples-to-apples" comparison is therefore bound to fail. Of course one could try to change the filtering options in the driver to find a middle ground, but that would always penalize one of the cards, since this "middle ground" doesn't exist in practice. It's just the way it is. ATi and NVIDIA use different texture filtering methods. Period. On top of all this, the quality can differ greatly from one application to the next, thanks to a sneaky little feature called application detection.

So what is a graphics card tester to do? We can still compare the different methods to each and show them next to the results of Microsoft's reference rasterizer. This is a good starting point for making visible the idiosyncrasies of the Radeon X800 and the GeForce 6800 Ultra. Of course, Refrast doesn't necessarily render an image that is 100% accurate, but it should be sufficient to get a good impression.

The following screenshots were taken in the Image Quality test of 3DMark 2003. This application is not only an excellent tool for testing image quality. As a very popular benchmark, it is also a prime "target" for especially aggressive optimizations to improve speed...

The images show a view into a tunnel, with the viewing plane being just below the middle. First we'll show you the results with standard trilinear filtering, once with a special black and white texture to show the practical result and once with once with colored mipmaps to show the angle-based optimizations of the drivers. In addition, there is also an image that shows the differences to Microsoft's reference rasterizer. All screenshots were taken at a resolution of 1024x768. (The NVIDIA driver used is 60.72)