After almost three happy years with the GTX 670 4GB SC in SLI on a 3770K, I decided to hand that down this week and build a new system around dual 970 FTW+. However, the performance increase in SLI does not seem to be holding up to what we came to expect in 670s. At least, not for me.
 
In reviews, I have seen claims of benchmarks showing the second 970 offers ~92% improvement. I'm seeing a 68% improvement. This seems weak.
 
The last tests I performed were 3DMark Fire Strike Extreme:
970 -- 5314
SLI -- 8946
SLI (10,365 graphics, 12,338 physics, 3669 combined).
 
On Unigine Heaven, I saw slightly better, but a huge drop in min fps in SLI:
970 -- 1306   51.8   25   117.9
SLI -- 2296   91.1   8.3   151.5
 
The highest VRAM usage during Unigine was 1201mb.
 
4790k
16GB 2133mhz @ 1600mhz
Asus Maximus VII Hero
Seasonic X-1050
Windows 7
Only drivers installed are latest NVIDIA and Intel's drivers for USB, SATA, and LAN.
Asus Swift PG278Qin gsync @ 144hz
I'm open to any input.

Regards. 

I just ran another set at Ultra (but with AA off) and the scores on first and second/third run were identical (2252 score, 89 fps, 36min 151 max) except for the first run which was 8.3 minimum again. This is at 2560x1440, but.. it still seems low, right? Especially over the speed of just a single card? I considered I might just have been spoiled by the 670s, but since I keep reading everyone (reviewers, that is) claiming they get almost double out of a second card, I feel like I am operating under expectation.
 
Also, these latest runs were after clocking my ram up to its 2133mhz (something I'm still unsure I want to do long term, because I'm not sure how safe 1.65v (and slightly above, according to the bios) is. I believe that's the high limit of Haswell, so... I suspected that since the 2133mhz runs at 9 or 10 CAS but at 11 or 12 at 1600mhz, increasing it might actually have some impact, but doesn't appear to be involved.

50-90% scaling, higher in games and benchmarks that are well optimized.