I think naming is GTX780 is a horrible idea. That’s just going to perpetuate the cycle of rebrading generations which aren’t deserving of the name. Like GTX580 should have been GTX485 since it was essentially a full working Fermi chip. Similarly, since GK110 is still a Kepler generation and won’t be 50-75% faster than GTX680, it shouldn’t be called GTX780 or that’s simply misleading. 

Nvidia starting the naming screw ups with GeForce 9 rebadges. It’s time they stop and go back to the old days of making sense when renaming their chips.
The estimate of 20-25% seems too low though. Based on 2304 SPs and much wider memory bandwidth it should scale better than that unless the card is clocked way below 1000mhz. 

It’s possible those are just preliminary estimates right now. Also, the chip may have less aggressive GPU Boost/Base clocks to fit within a certain TDP limit (250W). If it has a large overclocking headroom like Kepler, it could have more potential. 

Still even with 25% performance (which I admit I expected this to be more like 35-40%), it would be at least 35% faster than HD7970. That’s not bad if it launches in September/October, especially if NV lowers the price of GTX680 to $399 and positions GTX685 at $549. 

Interesting how in a span of less than 10 months, an HD7950 for $450 will look downright mid-range. Shows how seriously overpriced HD7900 series was at launch. 

25% performance boost- seems ok. To be honest i expected gk110 to be around 30-40% faster than gtx 680. I am not sure about should i upgrade or not. I currently have gtx 570. 680 did not push too far away, so i was hoping that gk 110 will get the job done. I am kinda of a guy who needs more raw performance( new features such as txaa and dynamic vsync are welcomed; lol). It all comes to 1 thing: how much will 685 cost? If it will be 550$ and then it is gonna rock, if more-:(  Do you guys think i should wait till gtx 685, upgrade my PC with 2nd 570 or get 780 or 770 in 20…

Thanks,
SOL

 That depends if you wish to wait few months for: a) GTX 685, b) GTX 680 price drop. If those rumors are true, then I would grab the cheapest GTX 680 and overclock it to its limits.

You should upgrade if:

1) The games you play no longer run to your satisfaction (FPS, min frames, image quality, etc.);
2) If #1 does not apply and you just have $ to burn;
3) If you need to keep up the e-peen on GPU forums.Otherwise, the general rule applies: The longer you wait, the more performance you’ll be able to buy at the same price level.

Waiting forever also doesn’t make sense. That’s why I always refer to rule #1. Personally, since videocards are the most depreciating assets in a modern desktop, I always use rule #1. If I have extra $ to burn, I just invest it into stocks, bonds, etc. And when Rule #1 applies, that’s when I upgrade. I think GTX570 with a mild overclock is plenty fast for 1080P. If you are gaming at 1600P, I’d upgrade. Otherwise, there is no point, especially since all the best games in 2012 are NOT GPU intensive (Starcraft 2 expansion, Diablo 3, Max Payne 3, Mass Effect 3, Bioshock Infinite, etc.). 2012 is really not a year worth upgrading for since there is no game like BF3, or Crysis or Far Cry, etc. It’s actually the best year to wait imho since 28nm process is very expensive and well the summer is here to enjoy outdoor activities > games!

My guess is Maxwell will be delayed until 2014 at the earliest. 

1) Previous timetable misses.

Nvidia has been 6 months late with Fermi and Kepler. Using that track record, I am going to say they’ll be also 6 months late with Kepler (so it’s reasonable to shift from 2nd half of 2013 to 1st half of 2014).2) Unexpectedly rising costs of node shrinks. 

When NV originally laid out the roadmap for Fermi (2009), Kepler (2011) and Maxwell (2013), they likely didn’t foresee how much more expensive the transitions to lower nodes will be AND how much more difficult they will be due to capacity constraints at TSMC. By the looks of it, NV will take another quarter to roll-out the full GTX600 series desktop line-up.  Since current GPU generations tend to last 18-24 months (Fermi came out March 26, 2010), it would only be half way through a generation to release Maxwell next year. 

3) Die size and TDP limitation of 28nm.

GK110 is supposed to be an enormous 500mm^2 die (or so) on 28nm process. That will push 250W+ TDP.  Considering Kepler architecture is the most efficient per transistor, per die, per watt already, it seems doubtful that Maxwell can improve on that in just 12 months from today. My best guess is Maxwell will launch on 20nm, not on 28nm. This against points to 2014 at the earliest.

4) Enormous costs to launch new architectures.

1 year of Kepler is not long enough to recoup the enormous R&D costs of 3-4 years to get that architecture running. 

5) Lack of strong competitive pressure.

If GK110 is 25% faster than GTX680, it would already be fast enough to compete against Sea Islands. Right now HD7970 needs similar clocks and 2048 SPs to match a 1536 SP GTX680. Unless Sea Islands has massive GPU clock speed increase, very large rebalancing of the architecture/massive increases in performance/clock, then it will take a 3000 SP Sea Islands to go against a 2304 SP Kepler.

There probably gonna release it as the 780

I wonder if this new card being pushed out so fast has anything to do with the RSOD?