There are very few cases where you can even show a difference for having 3GB. No games are pushing the kind of memory required for this. As soon as you push far enough to tap them out where it REALLY would make a difference you are running below 30fps a LOT and require TWO cards (1440p and up generally causes the 30fps problem which forces a 2 card situation where memory again doesn't matter yet). These cards are all tapping out the gpu itself at 1920x1200 or less in most games (not the memory), so the second you go above that you need the 2nd card which solves any memory issues anyway (since the 2nd comes with another 2GB or 3GB etc). Titan with 6GB to me is a waste (rather have superclocked 4GB memory). But then I guess that's why we have a 780 now with 3GB
Which again usually shows nothing, but I'd rather have 3GB over 2GB just in case it is handy one day even though I think you'll be under 30fps to prove it again.
If anyone has a website with benchmarks showing a SINGLE card making a REAL difference with anything over 2GB please give links
Every time I see a 2GB card vs. 3GB or 4GB of the same model I see NOTHING until they are BOTH below 30fps and require a 2nd card to make any difference. This doesn't count using them with APPS that use huge data sets though (I'm talking GAMING!). Pro-Apps can make a difference easily seen in quite a few situations. Games will get there when they actually start making games aimed at 3GB+. When 3GB is on the bottom rung, maybe they'll start making crap for it. You have to understand they are usually aiming for at worst, least common denominator, and at best the middle of the road. Only a handful of games on the market completely aim high end (crysis3?...LOL) and even they run into the previously mentioned 2card required problem (or 2 gpu on one card - 690, 7990 etc).
Again if you can point to 2GB showing worse than the same card in 3GB or 4GB where you are ABOVE 30fps min (or say 60fps avg - which are basically the same thing) then please give links.
With maxwell or volcanic islands we may get a gpu that can change this, where there will be more situations where the gpu itself can actually push 30fps min at 1440p etc a lot more so this may not need a 2nd card to be useful but until I see these there really isn't a point over 2GB. Of course if you can get an extra gig at about the same price take it. But if it's quite a bit more realize you're missing NOTHING, take the cheap card
Also realize according to steampowered.com survey only 1.25% of use use ABOVE 1920x1200 (and according to the same survey 99% of that 1.25% use 2 cards or more...LOL). This is why I think anandtech's recent 1440p articles are so dumb and their cpu recommendations at 1440p are ridiculous. An A8-5600 is fine for ANY single card they say...ROFL...Yeah if you're retarded and haven't seen this cpu benched vs an ivy bridge/haswell at 1920x1200 or 1920x1080 where it leaves AMD chips completely in the dust. They should be ashamed of making such statements. The cpus part like the red sea at 1920x1200 or below WITH A SINGLE CARD and do it again at above this RESOLUTION the second you add another (very needed) gpu, and this happens in FAR MORE than just civilization 5 as Ian Cutress says at anandtech. I don't know how much AMD is paying them to put crap like that out (but they were the only site visited personally by AMD recently...LOL - do the math - No FCAT etc there either - constant excuses for leaving that data out). I'm starting to tell friends in IT to avoid all their cpu or gpu articles as they are so slanted. Their alexa traffic is 1/2 off since their 660TI article in Sept or so, so clearly people are starting to see this and comments on those articles show people aren't fooled. You can only lie for so long before your reputation turns to crap.
Read the comments on the two 1440p articles
Or just read other sites, like this one, hardocp, techreport, pcper, etc etc...