davemaster84 :
Ehh I wouldn't call it a "stupid" thread, actually the ones that stated such kind of thing was Tom's reviewers, not to mention what they did with the 670 review (basically they buried the 680) . So if you think this kind of comparisons are stupid please do the same with the reviews Tom's published.
I do. Not necessarily with Toms but with a lot of reviewing outlets in general.
When the 7970 came out AMD didn't have full driver support for it. It was released in January with Catalyst 12.1, but full support wasn't available until Catalyst 12.3. When the reviewing outlets got their hands on it they ran it with the 12.1 drivers which is fully understandable because those were the ones available at the time. Since the drivers lacked full support, the benchmarks showed reasonable gains over the old HD 6970 and only marginal gains over the long-time flagship GTX 580.
When the 680 came out, the reviewers were were all too eager to see how it stacked up against the 7970. All too eager in fact that they didn't even bother to rerun the 7970 benchmarks with the new drivers, they just slapped the 680 benchmarks beside their existing 7970 benchmarks and yelled "first!".
Similarly, some reviewing outlets run benchmarks with the lowest common denominator as far as settings go. Most outlets don't test Metro 2033 with the Depth of Field filter turned on. Metro's DoF filter is Direct Compute based rather than shader based. This means that it runs exceptionally well on AMD cards but crushes NVidia's Kepler cards.
There's also variance in the set of applications that they benchmark against. A few reviewing outlets have a suite of applications that are almost exclusively TWIMTBP which means NVidia bought the exclusive rights to hand optimize them before they went gold. This means that AMD can't touch the game until it's been released while NVidia has had some time already. Even Tom's Hardware has been criticized for having a degree of bias in this regard.