Elite Bastards has made a new article in which they further investigate the latest suspected 'driver optimizations' in ATi his drivers. In this article they check out Aquamark 3 :
Luckily, since that initial article, we now have the information available to answer both of those questions. First off, let's tackle the simple question - Which image is the closest to being 'correct'?
As I mentioned in the first article, the most obvious way to check which image is closer to what the developer intended is to use the DirectX 9 reference rasteriser. Thankfully, since my initial article Massive Development (the developers of AquaMark 3) have stepped forward and very kindly made available to me an entire library of reference rasteriser images from their benchmark to use as a comparison. Many thanks go out to Massive for taking this step, and as such making a long-winded task much simpler.
According to Elite Bastards the found irregularities in Aquamark 3 are caused by a driver bug which possibly affects gamma levels so that it makes things darker than they should be.
There seems to be no performance benefit from doing this, so to call it a 'questionable optimisation' seems wholly inappropriate in light of this. It's a bug, pure and simple, and one that ATi will hopefully give a high priority in resolving. Of course, we are still waiting for word from ATi themselves about the issues brought up by Tom's Hardware, so hopefully once we have that the case will be closed... Until the next time one IHV decides to reveal something a competitor is rendering incorrectly at least.