I'll try one more time: it doesn't matter if it was an "old" test or not because the problems that may have existed with the card and/or the rendering have been fixed, so any newer results will only be equal or better than those. I posted that link because it used all 4 cards in the test and was easy to compare them against each other.
You're trying to cloud the issue with non-issues (as usual), but the fact remains that the 2900XT can run with the GTX. You should accept it and move on.
You don't seem to understand that having rendering problems will inflate scores, which is why those issues make results invalid.
To put it simply, if a square, rectangle and sphere are all supposed to be rendered but only the square and sphere are rendered, it will do so faster than having to render all three.
So comparing a card that renders 3 vs 2 and saying its a fair test is just bs.
Just because in the future the problem is fixed and now all 3 objects are rendered doesn't mean you can use the data where only 2 were rendered.
CoJ is another benchmark which was inflated because AF was not being applied yet many reviewers failed to comment on that and listed their results as having AF on, thus inflating the scores.