Iím sure youíve read reports around the web about Nvidia allegedly cheATing with their yet to be released 60.72 build drivers. These drivers were sent out to journalists evaluating the new GeForce 6800 Ultra and the NV4x architecture. Just two short weeks after the first reviews were posted one website published a lengthy article
discussing image quality and how some frames rendered by the new GeForce showed differences from the same frames rendered on ATi hardware. So is this the proverbial smoking gun weíre looking at, is Nvidia indeed making shortcuts to boost the performance of their new architecture or is something else going on here?
I think the jury is still out on what is the proper way to render frames; both companies have a different approach to calculating textures, doing anti-aliasing and all the other 3D calculations. Thus it comes as no surprise that Nvidia indeed renders scenes differently than ATi does, but where do you draw the line between a cheat, an optimization, a shortcut or an architectural difference? Both companies have used shortcuts before that some consider cheats and others a good way to boost performance. If a shortcut is only visible after enlarging portions of a screenshot to 4x its size, who really cares, as it is not visible during normal gameplay, and it does improve the performance.
Itís a matter of what shortcuts you use where, and both ATi and Nvidia have plenty of shortcuts built into their architecture. In essence for both companies to render the frame exactly the same, the same architectures would eventually be needed. However what makes graphics accelerators interesting, and gives us something to write about, is the way in which each company uses their own technology to deviate from this path to increase performance whilst producing the same result. Both ATi and Nvidia use proprietary technology to increase performance and efficiency which basically are shortcuts that all produce the same end result; the frame rendered.
Using an architecture specific feature to up the performance is a shortcut, so is rendering only the pixels that are visible in the frame. If these shortcuts donít decrease image quality, or only by a margin that is imperceptible, it surely is a legitimate shortcut, since it is producing the same end result. Following that train of thought, I always like to compare these discussions about image quality to MP3 compression. Millions of people use MP3s and don't hear the difference between the MP3 and the original CD. However if you look at it from a sound quality point of view there's a big difference, with the MP3 simply leaving out details you don't really hear at all. I have a similar approach to rendering quality, why render every pixel fully, if by using shortcuts to render just those that matter gives you better performance and identical image quality unless you scrutinize the screenshots with a microscope.