ATI's Optimized Texture Filtering Called Into Question


All Filter optimizations discussed here aim to increase the performance of the graphics cards without materially reducing image quality. The word "materially" is, however, subjective - depending on the optimization used, a loss in quality is perceptible when taking a closer look. Even if the quality in screenshots is OK, a running game is often a different chapter. Annoying effects (moiré, flickering) can crop up that were not noticeable on screenshots.

In the case of graphics cards in the medium and lower price segment, the customer will certainly get added value in the filter optimizations, because "correct" filtering would slow the chips down too much. The user can play in higher resolutions or add filter effects that without the optimizations would be unplayable. The bottom line is that the customer ends up with better image quality.

It's a different story with the new enthusiast cards, such as the Radeon X800 Pro/XT and the GeForce 6800 Ultra/GT. With those cards the optimizations do not provide the customer with new added value - on the contrary. He gets a reduced image quality, although the card would actually be fast enough to deliver maximum quality at what would surely still be an excellent frame rate. We cannot escape the impression that the filter optimizations in the new top models will no longer be used ultimately to offer the customer added value, but rather solely in order to beat the competition in the benchmark tables, which are so important in the prestige category. Whether or not the customer will be ready to spend $400-$500 for this is quite another matter. NVIDIA has obviously realized this and allows true trilinear filtering as an option in its newest models. Well, it did not work in the latest v61.11 beta driver because of a bug... let's hope it indeed is a bug and will work again in the final driver release.

"As long as this is achieved, there is no "right" or "wrong" way to implement the filtering..." (Thanks to .

To be continued? One would assume so, because the discussion about the recently discovered brilinear filtering in the Radeon 9600 and X800 is still going full steam ahead. ATi deserves credit for the fact that the image quality of the cards is not visibly compromised by this filtering; at least no example has yet been seen of this. So far, the brilinear areas have only showed up in laborious tests. However, ATi is currently not offering true trilinear filtering with the cards mentioned above, whether adaptive or not. Because of the new filtering, the performance values of the benchmarks do not show the true potential of the X800, because the FPS values only occur due to an optimization whose details are unknown. Even the word adaptive has a bitter aftertaste. ATi has not provided information about the way the driver works and has declared numerous times that it is offering true trilinear filtering. Only since the discovery was made has ATI admitted that the filtering is optimized. Hopefully this type of adaptivity is not being used in other places in the driver .

However, slowly but surely manufacturers are moving to the point where tolerable limits are being exceeded. "Adaptivity" or application detection prevent test applications from showing the real behavior of the card in games. The image quality in games can differ depending on the driver used or on the user. The manufacturers can therefore fiddle with the driver, depending on what performance marketing needs at a given moment. The customer's right to know what he is actually buying therefore falls by the wayside. All that is left for the media is to limp along with their educational mission. The filter tricks discussed in this article are only the well-known cases. How large the unknown quantity is cannot even be guessed.

Every manufacturer decides for itself what kind of image quality it will provide as a standard. It should, however, document the optimizations used, especially when they do not come to light in established tests, as lately seen with ATi. The solution is obvious: make it possible to switch off the optimizations. Then the customer can decide for himself where his added value lies - more FPS or maximum image quality. There is no real hope that Microsoft will act to police optimization. The WHQL tests fail to cover most of them and also can be easily evaded, read: adaptivity.

Swipe to scroll horizontally
Currently Known Filter Optimizations
Row 0 - Cell 0 ATINVIDIA
TrilinearoptimizationR9600X800GF FX5xxx(GF 6xxx)*
Angle optimizedanisotropic filteringR9xxxX800GF 6xxx
Adaptiveanisotropic filteringR9xxxX800GF FX5xxxGF 6xxx
Stage optimizationR9xxxX800GF FX5xxx
Optimized LOD-FractionR9xxxX800(?)Row 5 - Cell 2

Still, the ongoing discussion also has its benefits - the buyer, and perhaps, ultimately, OEMs are being sensitized to this issue. Because the irrepressible optimization mania will surely continue. However, there are also bright spots in the picture, as demonstrated by NVIDIA's trilinear optimization. We hope to see more of the same!