Here the latest rumor for the new Dual Chip GTX 900 series VGA card currently named GeFore GTX 990:
http://pokde.net/tech-blog/nvidia-will-launch-geforce-gtx-990-early-2015/
 
 
It is rumored to be launched early 2015
My Speculations would be a similar HSF like the GTX 690 & Titan Z have but Dual Slot with these specs:
 
2x GM204-430-B1
2x 2048 Maxwell CUDA Cores @ GTX 970 Speeds
2x 4096MB 256Bit GDDR5 @ GTX 970 Speeds
275 a 300 Watt TDP
1x 8pins + 1x 6pins for Power input, just like GTX 295 has.
 
Just my guess of the card, I could be far off, I could be close
Got more rumors ornews of this new GX2 Style 980? Feel free to post more
 
 

GTX980 is effectively similar to GTX680 in power even without 20nm, it more than allows a GTX690 successor in the same envelope (dual-slot, 8+6 pin, 300W or less) apart from the advantages due to Maxwell.  I don't think GM210 needs 20nm, since nVidia could just introduce another 250W card.   
 
It's AMD that needs 20nm for their graphics more than nVidia, though I read a rumor that they'll be skipping 20nm for their APUs and going straight to 16nm FinFET production-- catching up to Broadwell.  Next year will be very interesting for both GPU and CPU.

I just feel it happening, I even want to use my good ole GTX 295 Red Editions in Quad SLI, just see what I can get running with Dem xDDD But But I I want two GTX 990's so badly cause well youall remember my 3dfx Voodoo5 6000 days right Quad SLI...Hmmm Quuaaad haha still wuvs its xDDDDDDD Arrr Rarr Rawr Dwooollzz xDDD

Semiroundboss
I see the 990 taking cues from the R9 295X2. Meaning that it will be two 980's overclocked instead of two 980's underclocked or at reference speeds like previous dual gpu cards from Nvidia have been. If this card is 16gb, 16nm or 20nm like rumors have suggested, and faster than two reference 980's, it will dominate for enthusiast PC's. The only thing that could touch the 990 from team green, if it is in fact 16gb and 16nm, would be if Asus made an ROG Mars 970X2, which I don't see Nvidia allowing, since it would cut 990 sales in half the same way a 6gb 780 Ti would have to the Titan Black. It will be interesting to see what team red does to combat the 990.

Hmm Interesting idea's you have there
 
Although I doubt NV giving this card 2x 8GB 256Bit GDDR5, 2x 4GB 256Bit GDDR5 would be more realistic, remember with GTX 690 NV also only went for the default VRAM Size a GTX 680 had which was 2GB over 4GB 256Bit GDDR5?
 
However ASUS did make a few Engineering Samples of their never released ASUS MARS III which was a GTX 690 @ GTX 680 Speeds and  with 2x 4GB 256Bit GDDR5 as well, o I only think ASUS will make such crazy over sized designs, even though I find it stupid of NVIDIA to not let a 3rd party company make their design better, they will sell more GPU's this way likewise and save them money for making boards themselves

candle_86

Gold Leader

Semiroundboss
I see the 990 taking cues from the R9 295X2. Meaning that it will be two 980's overclocked instead of two 980's underclocked or at reference speeds like previous dual gpu cards from Nvidia have been. If this card is 16gb, 16nm or 20nm like rumors have suggested, and faster than two reference 980's, it will dominate for enthusiast PC's. The only thing that could touch the 990 from team green, if it is in fact 16gb and 16nm, would be if Asus made an ROG Mars 970X2, which I don't see Nvidia allowing, since it would cut 990 sales in half the same way a 6gb 780 Ti would have to the Titan Black. It will be interesting to see what team red does to combat the 990.

Hmm Interesting idea's you have there
 
Although I doubt NV giving this card 2x 8GB 256Bit GDDR5, 2x 4GB 256Bit GDDR5 would be more realistic, remember with GTX 690 NV also only went for the default VRAM Size a GTX 680 had which was 2GB over 4GB 256Bit GDDR5?
 
However ASUS did make a few Engineering Samples of their never released ASUS MARS III which was a GTX 690 @ GTX 680 Speeds and  with 2x 4GB 256Bit GDDR5 as well, o I only think ASUS will make such crazy over sized designs, even though I find it stupid of NVIDIA to not let a 3rd party company make their design better, they will sell more GPU's this way likewise and save them money for making boards themselves

hey I want them to, I still remember the 7800GT Dual card

Matter of fact i still want one lol

I got into PC gaming around the time the 600 series was coming to an end, so I'd never seen that card before. But that looks like it would have been insane for the time!

Gold Leader

Semiroundboss
I see the 990 taking cues from the R9 295X2. Meaning that it will be two 980's overclocked instead of two 980's underclocked or at reference speeds like previous dual gpu cards from Nvidia have been. If this card is 16gb, 16nm or 20nm like rumors have suggested, and faster than two reference 980's, it will dominate for enthusiast PC's. The only thing that could touch the 990 from team green, if it is in fact 16gb and 16nm, would be if Asus made an ROG Mars 970X2, which I don't see Nvidia allowing, since it would cut 990 sales in half the same way a 6gb 780 Ti would have to the Titan Black. It will be interesting to see what team red does to combat the 990.

Hmm Interesting idea's you have there
 
Although I doubt NV giving this card 2x 8GB 256Bit GDDR5, 2x 4GB 256Bit GDDR5 would be more realistic, remember with GTX 690 NV also only went for the default VRAM Size a GTX 680 had which was 2GB over 4GB 256Bit GDDR5?
 
However ASUS did make a few Engineering Samples of their never released ASUS MARS III which was a GTX 690 @ GTX 680 Speeds and  with 2x 4GB 256Bit GDDR5 as well, o I only think ASUS will make such crazy over sized designs, even though I find it stupid of NVIDIA to not let a 3rd party company make their design better, they will sell more GPU's this way likewise and save them money for making boards themselves

That's weird Gold, Asus released the Mars III as a custom dual 760 card:
 
http://rog.asus.com/tag/mars-iii/
 
Newegg even has one for sale:
 
http://www.newegg.com/Product/Product.aspx?Item=N82E16814121839
 

Evga at one point had a dual680chip card they showed at a trade show. Pics of it leaked and the mods spent the next few weeks freaking out and annoying everyone by deleting every thread that mentioned it. Kinda sad it never went into production. Did seem a bit large tho it was quad slot.

Bruno747
Evga at one point had a dual680chip card they showed at a trade show. Pics of it leaked and the mods spent the next few weeks freaking out and annoying everyone by deleting every thread that mentioned it. Kinda sad it never went into production. Did seem a bit large tho it was quad slot.

I thought that the GTX 690 had Two 680 GPU's in it?

would be nice, I'm waiting to see a 960 GTX Ti...I can hope...

Squall_Rinoa86
would be nice, I'm waiting to see a 960 GTX Ti...I can hope...

Why do you think that NVIDIA will come out with a GTX 960Ti when they never came out with a GTX 760Ti other than the OEM one?
And was even rated lower then the GTX 760

Bruno747
Evga at one point had a dual680chip card they showed at a trade show. Pics of it leaked and the mods spent the next few weeks freaking out and annoying everyone by deleting every thread that mentioned it. Kinda sad it never went into production. Did seem a bit large tho it was quad slot.

That was the GTX 680 2Win Gemini that never saw the day of light but just some trade shows
 
http://www.techpowerup.com/gpudb/b1124/evga-gtx-680-2win-gemini.html
 
Don't know why even the EVGA staff doesn't talk about that thing.

dual gpu and only uses "1x 8pins + 1x 6pins for Power input".
 
how is that even possible?

rcarlos243
dual gpu and only uses "1x 8pins + 1x 6pins for Power input".
how is that even possible?

Dual GPU cards share a lot of circuitry, the other stuff on the PCB than just the GPU and VRAM.  OEMs typically add more of that circuitry on single-GPU high-end to better sell/ensure the potential beyond stock speeds-- but there isn't enough room for all that on a dual-GPU card (although ASUS has been known to do so anyway on their ARES cards which result in needing more than one of each power plug).  Furthermore, the clockspeeds are typically lowered on reference dual-GPU cards.  
 
 
Lastly, desktop graphics cards aren't as energy efficient as you might think, they are simply the cheapest they can be while being massively produced.  Say you wanted to keep the features and make something smaller, that adds cost.  If you wanted to make it use less energy while keeping it smaller, that adds even more cost.  It explains why high-end laptops are so expensive, even though their graphics card tend to be the equivalent to midrange desktop.
 
For instance, the mobile GTX980M card costs $720 for what is around the same performance as a single GTX970 desktop card in FireStrike scores (GTX970 versus GTX980M), but uses half the power and is half the size.  If a GTX990 used two of these instead, making the card cost $999 is a no-brainer, in fact is probably too cheap.  Except each GTX980M doesn't even use more than 100W each, so two would be around 200W (needing only 2x 6-pin or 1x 8-pin) -- but more than likely the desktop GTX990 will be closer to 300W due to higher clockspeeds than the mobile card with its 5GHz Vram.
 
In my mind it makes more sense to use a pair of the GM204 with 4 SMM clusters disabled than to use a full GM204 die, uses less power and get to the point of getting GTX970 SLI in a single package (because that was the same case before where GTX690 = 2-way SLI GTX670).
 

Gold Leader
Uhm GTX 690 was GTX 680 SLI each GK104-400-A1 has 1536 Kepelr CUDA cores and not 1344 which the GTX 670's had...

I know, but the GTX690 still performed like a pair of GTX670 due to the lower frequencies on those 1536-CUDA; we can't just focus all predictions on the CUDA cores alone.  Titan-Z, in fact, same total CUDA and memory bandwidth per GPU as a pair of Titan Black, but the GPU frequencies were low enough enough to match up to a 2-way SLI GTX780 with less CUDA and memory bandwidth.  When people don't upgrade often, they need a reference and thus looking for performance equivalencies, not exact specs.
 
I'm just putting the idea forward that using the dies from a mobile GTX980M at reference clocks rather than a pair of downclocked desktop GTX980 will make a dual-GPU card use less power and perform similar to a pair of desktop GTX970.  Both dies are still GM204, but different revisions.

Rei86
 
Not only that being on the same PCB caused quite a bit of issues on the power/cooling front. 

I'd attribute that to heatsink + fan + shroud design.  Not that the older dual-GPU setups were favorable before a single-PCB married both GPUs, but the current design since GTX295 has a fan between both GPUs and half the heat recirculated in the system.  Even 9800Gx2 vented most of the heat to the side, which depending on the chassis, wasn't favorable either.  In fact, looking back not even 7950Gx2 was best with a pair of coolers tat still vented back into the chassis.
 
 
I remember ATI/AMD had a different approach early (HD3870x2, HD4870x2, HD5970), using a regular blower configuration to vent outside the chassis, but the first GPU exhaust went over the second.  
 
Seems the only way to circumvent the heat issue is to isolate the heat away from the system and not affect the other GPU, other than merging both dual GPUs close enough on the PCB to share the same heatsink, is to try liquid like R9-295X2.