The window for making the SLI upgrade generally lasts 18 - 24 months. We're just past that.
Two 970s was a proverbial no brainer as it was cheaper and outperformed the 980 by some 40%. That number has been historically valid based upon TPUs 19 game test suite, but it's not like it is carved in stone. You will get better scaling in really demanding games, now sometimes in excess of 100% even; but on games already getting 90+ fps the scaling is not so good (not that you would really care). Do you buy an x80 series card over a x70 to take you from 90 to 108 fps ?... or do you buy it to take you from 35 to 42 fps ? If it's the later, why not two x70s to get 60 fps ? ... and yes those values are proportionally accurate.
Knocks against average scaling are therefore simply a huge red herring. With 970s for example ...
Tomb Raider goes from 29.8 to 58.7; scaling = 96.98%
Battlefield 3 goes from 62.1 to 121.4; scaling = 95.49%
Far Cry 3 goes from 35.6 to 68.8; scaling = 93.26%
Crysis 3 goes from 22.5 to 43.3; scaling = 92.44%
Thief goes from 70.8 to 136.1; scaling = 92.23%
Bioshock Infinite goes from 76.7 to 143.9; scaling = 87.61%
Splinter Cell: Blacklist goes from 49.5 to 92.2; scaling = 86.26%
Battlefield 4 goes from 45.0 to 83.2; scaling =84.89%
Metro LL goes from 40.7 to 74.6; scaling =83.29%
Batman: Arkham Origins goes from 81.8 to 148.3; scaling = 81.30%
Yes, Alien Isolation goes from 87.3 to 121.0; scaling = 38.60% ... Do you really care ?
Tom Raider went from a disastrous < 30 fps to 60 fps territory with twin 970s, that has an impact;
do you want the 58.7 fps of SLI'd 970s of the 35 fps of a single 980 ?
Crysis 3 went from a disastrous 22.5 fps to a almost satisfactory 43.3 w/ twin 970s .. whaddya like better , the 43.3 of the twin 970s or the 26.8 fps of the 980 ?
Far Cry 3 went from 35.6 fps to a quite satisfactory 68.8 with twin 970s; whaddya like better , the 68.8 of the twin 970s or the 41 fps of the 980 ?
Alien isolation only went up 39% but since you were already getting well above 60 fps, it has no impact.
As evidenced by the above, I see little support for the claim that the single card is "almost always better". Where it mattered, the twin 970s were 66% faster than the 980 for the same price. Where SLI scaling is poor, the 2nd 970 may not be doing much for you but then again neither is the 980....you could generally get by with a 960 or 950 in these instances.
However, generalities aside, we still have a card that is more than 2 years old where SLI's advantages wane quickly. And, at this point, we just don't know if the 1070 will continue that same SLI advantage over the single 1080 so it's a question that can not be answered as yet. It's further complicated by the relative overclocking ability of the cards... a good OC on a 970 was 17% ... a good OC on a 980 was 25% ... a good OC on a 980 Ti was 30%. So if two 970s had an advantage of 40% at stock settings, that advantage dropped to 37% when all were overclocked.
Our problem at this point is that no one's really allowed to talk about the non-reference 970s till Friday ... and I don't think we'll see a fair number of reviews including SLI for a week or so ... and that's **if** the various glitches that were seen in the 1080 SLI test were fixed by yesterday's driver
Speaking specifically to the cards in question, The 1080 is 2.4 times as fast (all on average) as the 780 and the 1070 is about 80% as fast as the 1080 ... so roughly, the 1070 is 1.9 times as fast as the 780
Two 780s in SLI are 1.7 times as fast as a single 780 ... so we are left with the 1070 being about 12% faster than twin 780s
Factors ...
1. The non reference 1070 will likely cost you $399 ... the 780 $200. So is it worth twice as much to get 12% more speed ? Can you sell the 780 for $200 ... if ya can, then the 1070 at a net cost of $200 is a no brainer.
2. Power and heat... Measured power (average under max load) is 161 for the FE 1070... figure 175 - 180 for the non -reference jobs.... versus about 500 watts for the twin 780s.
Give the above, ya gotta like the 1070 over the twin 780s... of course, it loses a bit of attractiveness if ya can't offload the existing 780. However, w/ a 1200 watt PSU... you're wasting a helluva lotta energy and making a nice bit of heat
http://www.tomshardware.com/reviews/power-supply-psu-review,2916-3.html