[citation][nom]qiplayer[/nom]Gtx680 uses less power than the 7970. If you oc both at the maximum, lets say to 1250/1300mhz the difference is even more visible. A 7970 windforce +system had a powerdraw of 570watts, where without oc it was 430 about.This to say the 680 is far a better card, nvidia has yo struggle less also with cooling, and has had an easier job that amd will, in making dual gpu's.If we look at games it has sense imho to take the card that runs better the game you play.For the computing I think it's just not activated for marketing reasons. But I don't know gpu engeneering.[/citation]
Nvidia used a smaller die and sacrificed compute performance for that power efficiency. 680 owners had better hope that games using DP math don't get common for years or else they will have crap performance in them compared to the 7970. Let's not forget how the 680 loses power efficiency at high clocks more than the 7970 does.
[citation][nom]outlw6669[/nom]Actually, the GK104 is nVidia's mainstream chip and was not designed with compute performance in mind.Their actual highend chip, the GK110, is still to be released and was designed with a focus on compute.Dropping all those extra transistors needed for compute allowed nVidia to build a gaming chip that is smaller and more efficient (for gaming) than ATI's 7xxx series.The past few generations nVidia has been pushing hard compute performance at a sacrifice to gaming performance, die size and efficiency while AMD focused on gaming performance while sacrificing compute performance.This generation they both decided to switch roles and, as AMD is unwilling to outspend nVidia on die size (GF100 and GF110 where, while GK110 will probably be, 520mm²+ behemoths), they unfortunately ended up with a marginally inferior gaming chip that is massively better with computing tasks.[/citation]
GK104 and GK110? So, what happened to GK100? Does the Big Kepler skip a generation? That's unheard of and undoubtedly wrong. GK100 is the next big die size. GK110 would mean it's a second generation Big Kepler chip, yet there isn't even a first generation Big Kepler chip yet, so how is that possible?
Otherwise, agreed.