NV40, you should read the thread at beyond3d about that article. They raise some very good points. For instance, the reference image is not valid for pixel comparisons across different rendering methods of different architectures. Moreover, the difference between the reference image and an IHV's output can be caused by varying precisions. The CPU has been suggested to be using lower precision.
The style of the article was very unprofessional and sensationalist.
Another edit: If it turned out that ATI's lack of 32FPP caused the difference, then that's not cheating, just a drawback of the architecture.
Last edited by extreme_dB; 07-09-03 at 08:31 PM.