hm that is interesting, tomshardware says its slower then the 7870 (slightly) but here and anandtech say its 10% faster. I dont even know.
Any individual sites conclusions are going to depend on the benchmark suite employed, level of game I.Q. and resolution. The 660Ti's performance is affected by high levels of AA and image quality- increase both and the 7870 is definitely going to assert itself. The 660Ti's 192-bit bus width won't help it at higher screen resolutions either. To get an accurate -usable- picture you would need to factor in the likely system set up that you/likely target market would employ + benchmarking philosophy of the reviewers involved. I'd tend to note:
-Whether image quality trumps gameplay framerate- an example being DiRT Showdown on "Ultra" (also in line with
Hexus's figures) which would suggest testing largely for academic interest since I doubt anyone would in actuality suffer through low framerates (30-45) rather than ease up on game image quality and get a more fluid gaming experience.
-Whether the game selection is relevant within any given benchmark suite ( How current, popularity, weighted towards a particular genre)
-The screen resolution commensurate with the price point of the card(s)
-Whether the testing encompasses factors that are relevant ( I.e. max power draw/heat/noise under OCCT/Furmark for example)
Another thing to note would be whether the testing is "stock vs stock" situation - the GTX 660Ti may have reference clocks (although not many vendors are adhering to it), but I don't think there is a reference design, so many of these tests are in effect OC vs stock- although I note that a few sites are using AMD's new HD 7950 bios, which just adds to the level of subjectivity- since anyone competant enough to flash a VGA BIOS is generally competant enough to manually overclock using less voltage- and thus, less noise, heat output, and power consumption.
I think that the days of a straight comparison presented in a single set of performance figures are pretty much numbered for graphics reviews. The adoption of boost clocks, power saving configurations, adaptive v-sync (AMD will surely follow suit here) are rapidly marginalizing the traditional FPS bar graphs in favour of "gameplay-ability" - I.e.
fluidity of rendering (framerate latency) for example