After criticizing rival Nvidia for using optimized trilinear filtering algorithms in its PC graphics card drivers--calling them "unacceptable tricks" in a recent presentation to hardware reviewers last month--ATI Technologies has acknowledged using similar optimizations itself.
The graphics company's admission was a response to allegations raised by German Web site ComputerBase.de. ComputerBase.de staffers voiced concerns that ATI was possibly using application-specific cheats or optimized filtering algorithms after observing that the company's drivers rendered colored mip maps, such as those used in synthetic image-quality tests, with full trilinear filtering enabled, but the filtering method seemed to change under real game conditions in Unreal Tournament 2003.
For the layman, mip maps are a sequence of textures, each of which is a progressively lower-resolution representation of the same image. Mimicking real life, where people perceive less detail the farther they are from a subject, games use detailed mip maps in areas closest to the player's viewpoint while using less-detailed mip maps in more distant areas.
Trilinear filtering is a technique used to blend together two adjacent mip maps in the game environment to eliminate noticeable areas of "mip map banding" in the game where high-resolution mip maps transition to a lower level of detail. Trilinear filtering, however, is notoriously demanding in terms of graphics performance, but optimizations can be made to create similar image quality with less work.
In an online chat hosted by ATI to discuss the allegations, the company admitted that its graphics drivers do contain adaptive filtering optimizations but vehemently denied using application-specific cheats. Company representatives stated, "Our target is also to avoid any need to detect applications, and as such we have to try to be sure that our image quality remains high in all cases. To achieve this we spent a lot of effort developing algorithms to make the best use of our quality tuning options. This isnt a performance enhancement applied to popular gaming benchmarks."
When asked about trilinear optimizations in light of ATI's recent disclosure, Nvidia's Brian Burke commented, "In our view, if an optimization produces the correct image while speeding up performance, then it is beneficial to the end user and is legitimate. If a change in the driver does not produce the correct image, or functions only in the benchmark, it is either a bug and must be fixed, or a cheat."
ATI's algorithms appear to fit into the legitimate category with professional hardware Web sites hard-pressed to find any image-quality degradations caused by the filtering optimizations. According to Tech Report's Scott Wasson, "ATI's adaptive trilinear appears to be damn near impossible to catch with the naked eye."
With ATI's recent admission, consumers have another image-quality factor to compare since it is now apparent that both Nvidia and ATI sport different trilinear filtering algorithms that have their own specific trade-offs between performance and image quality.
While ATI has been cleared of cheating charges in the technical arena, the company still has to overcome the damage caused by its failure to disclose the existence of the optimizations to the editors of major hardware publications. Tech Report's Wasson has also reported on an ATI-issued Image Quality Guide where the company claims that "gamers get full trilinear...all of the time" with its newest Radeon X800 graphics card. While trilinear filtering can be broadly defined as any filtering technique used to eliminate the banding between mip map levels, it's commonly understood that "full trilinear" refers to classic, eight-sample trilinear filtering, but ATI has admitted that its graphics drivers only fall back to "legacy trilinear filtering" in situations where the texture content isn't applicable for its optimized filtering techniques.
Wasson goes on to point out that ATI recommended image-quality tools that use colored mip maps to reviewers and also advised users to disable Nvidia's trilinear optimizations while benchmarking to "ensure a fair comparison." ATI has not yet responded to GameSpot's inquires about the subject, but it appears that such a comparison isn't fair when ATI fails to disclose that its drivers are also running filtering optimizations that increase performance by "the order of a few percent" and are undetectable in colored mip map image-quality tests.
Reviewers have asked for ATI to include the option to disable the feature since filtering optimizations require subjective image-quality tests to gauge overall effectiveness, while classical trilinear testing is still useful as a raw performance benchmark. Nvidia encountered similar demands when it first introduced its trilinear optimizations and responded by adding the option to disable the feature within the driver menu settings. ATI has stated that it is open to the possibility of adding the option to future driver revisions but has not yet confirmed if or when such an option would be implemented.