homerdog :
I trust your quoted figures.
Unfortunately I have been trained to trust published benchmarks as much as I trust CNN and Fox News.
Benchmarks serve their purpose, but they are not an accurate representation of how pleasurable your gaming experience will be,
I find CNN and Fox to be worlds apart on opinion but there's really no difference in respect to what they report as fact. As far as I know, they have both been right about who won each election for governor in all 50 states, the guy they said won the presidential election does sit in the white house, and they each correctly reported who won the last world series as far back as I can remember.
A game benchmark is simply a "section" or a game or barometer of how long it takes to play that section. These sections are chosen to give as wide as possible set of conditions given that the benchmark must be limited to a reasonable length of time to complete. As such, if a particular card or driver particularly benefits from certain individual coding techniques, that game may put a slight weight to the card it favors. But, to my mind, a Crysis benchmark gives a pretty good indication of how fast a card will run Crysis, and AutoCAD benchmark gives a pretty good indication of how fast a card will run AutoCAD. So I wouldn't say that one benchmark test is something to base a selection on. But when ya compare two cards in 20 benchmarks ond one wins 19 outta 20, well to me, with a 19 to 1 advantage, I'd say it's pretty much a done deal as to which one is going to be fastest in most situations.
Of course, speed is not the only criteria to judge by. Back in the old days, I'd put Diamond cards in game machines and Matrox cards in multimedia machines as the Matrox cards had better color accuracy, but were generally slower than the Diamond's.....so if the main job was photo editing, the Matrox made more sense. CAD boxes on the other hand seemed to work best with gaming cards unless of course you were able to spring for one of the $2500+ imaging monsters. As for a certain web sites mandate that only subjective testing should be used, to me that just seems an excuse to play favorites.