I'm still running on an X58 motherboard with an i7-920 OC'd to 3.6Ghz, 12GB RAM, with an SSD (though relegated to SATA2 on this MB).  I'll be playing my share of Elite:Dangerous and Star Citizen so I thought I'd see if the CPU or GPU might be bottlenecking my system.  When I tested Elite:Dangerous with MSI's Afterburner and I saw that my CPUs utilization was running around 55% and the GPU's around 90% my decision was made.  It was upgrade time.  I very quickly found I wasn't going to locate another GTX 670 to run in SLI.  And, just as well as the added power consumption and heat wouldn't be welcome.  I knew I was going to stick with EVGA and went hunting up a GTX 980.  The GTX 980 SC ACX 2.0 was what I settled on.
 
While waiting for the upgrade to arrive I decided to run two benchmarks: Unigine Heaven 4.0 and 3dMark Firestrike.  I just installed the upgrade, ran the same benchmarks, and came here to report:
 
Unigine Heaven 4.0 with 4x AA (couldn't resist and always turn it on in my games):
GTX 670 FTW - FPS: 55.1 - Score: 1,388 - Temp: 77C
GTX 980 SC ACX 2.0 - FPS: 112.5 - Score - 2,834 - Temp: 69C
 
3DMark Firestrike (free version so no settings available):
GTX 670 FTW - Score: 5,930 - Graphics: 6,624 - Physics: 9,036 - Combined: 2,578
GTX 980 SC ACX 2.0 - Score: 10,857 - Graphics: 13,939 - Physics: 9,040 - Combined: 4,608
 
It's very interesting to see that the physics score is essentially the same between the two cards.  I love the drop in temps and the idea of drawing less power.  I can't speak to noise as I'm running no less than 6 fans on the case itself and 2 on the CPU cooler.  I've not yet re-run my test with MSI's Afterburner in a session of Elite:Dangerous.  Obviously, it's not a benchmark with a strict regiment of tests so it will be somewhat subjective.  But it should be very indicative of any change.
 
I don't know if this aids anyone who's looking at an upgrade or not.  But thought I'd share my experience.
 
Happy gaming and I love me some EVGA!

Indeed, the PhysX calculations were probably running on the CPU; you may be able to change that in the Nvidia control panel, though, and see what scores you get. Otherwise, you could run the 670 with the 980 as a dedicated PhysX card, then reverse the two, if you'd like. That could make for some interesting results.

I wonder how comparable two 670 FTW would be.  Looks like it would be very close.