I'm starting off with some early benches compared to my 580s. I'm not trying to go crazy yet with the new 680s, just showing off some potential.
First up are 3DMark11 and Heaven 3.0.
3DMark11 Benches It's going to flow like this: 580s@Stock vs 680s@Stock (Performance, Extreme), then rinse and repeat at overclocks.
EVGA GTX 580 SCs SLI - 797 Mhz 
Performance Preset 
Extreme Preset EVGA GTX 680s SLI - 1006 Mhz 
Performance Preset 
Extreme Preset Okay, so at stock clocks we see a 30% improvement at the Performance Preset and 42% improvement over the GTX 580s at the Extreme Preset at stock settings. That's pretty darn stout.
Now for OC settings.
EVGA GTX 580 SCs SLI - 950 Mhz 
Performance Preset 
Extreme Preset EVGA GTX 680s SLI - +100 Mhz Offset 
Performance Preset 
Extreme Preset This was a pretty substantial OC on the 580s and from all accounts a +100 clock offset on the 680s is about half what you can expect overall, so at those clocks we see a 24% improvement at the Performance Preset and 37% improvement over the GTX 580s at the Extreme Preset. That's awesome. Once I get these things warmed up and tuned in and the drivers mature some... P20K+ and X7K+ are well within reach.
Heaven 3.0 Benches Same flow as above but less screenshots since I only run one configuration here: 1920x1080 fullscreen, 16xAF, 8xAA, tessellation normal, and textures high.
EVGA GTX 580 SCs SLI - 797 Mhz 
EVGA GTX 680s SLI - 1006 Mhz 
At stock settings, the 680s hold a 23% gain over the 580s in SLI.
EVGA GTX 580 SCs SLI - 950 Mhz 
EVGA GTX 680s SLI - +100 Mhz Offset 
So again, we an improvement, but only about 13.6% at the OC clocks. This is due to me just being conversative on my 680s OC until I get used to all the new settings and managing the new Precision X which is really a giant step forward from Precision v2.x.
Power The value below is total system draw (excluding monitor) from the wall measured in AC watts. See below for more discussion.
3DMark11 at Stock Clocks 
3DMark11 at +100 Clock Offset with 120% Power Offset 
These were the highest I saw, typically values were in the mid 500s at stock and around 600 for OC.
So my PSU has about an 85% efficiency rating converting AC to DC. That means we're talking about roughly 516-568 watts DC after converting. These 680s offer great performance and far more economical on power than the 580s and the 480s before that.
Acoustics This is by far the most subjective and least accurate. I used my DB Meter app on my iPhone and at 85% fan I measured 71 dbA and at idle I still measured 61 dbA. So let's assume that the app is junk at accurate initial measurement of ambient noise but the delta between idle and 85% is accurate. A 10 dbA change isn't pretty good. That said, they sound just like the 580s, subjectively speaking so whatever you're used to there, expect it here with your custom fan profiles.
Battlefield 3 Okay, so we all play BF3. I personally love the Frostbite 2 engine. The good news? The 680s master it. My 580s in SLI would chew up BF3, but the 680s act like the Lindt chocolatier whisking his confection into perfection (if you've seen this commercial, you know what I mean). Where the 580s were a pair of rhinoceroses, these 680s were were more akin to any manner of large cat.
So here's a screenshot at round start, which was the only one I took because I got really into the round:
First, I had the cards overclocked by +100 core / +250 memory offset which yielded effective clocks of 1228 core / 6502 memory (up from 1006 / 6000). 105 fps was actually on the low side. I looked up often to monitor the the vitals and never observed anything below 100 fps and routine was in the 120-140 range. It wasn't a full server but I spent plenty of time in the action so I can say this cards were definitely crushing things in multiplayer. The temps represented above were a little low, they averaged 70/60*C for #1/#2 respectively at 85% fan. I was running 1920x1080 on the Ultra preset. I haven't tried FXAA yet in lieu of 4xMSAA, but I hear that could make it even better.
The Elder Scrolls V: Skyrim Okay, so I haven't really played Skyrim except for about half an hour. Shameful, I know. I need time to get into it and I just haven't had it so BF3 has been the go to game of late. That said, I fired it up anyway just to take a few screens. The good news is that Skyrim immediately detected new hardware and recommended 1920x1080 on Ultra-High settings. I then changed one thing only. FXAA was not checked, so I checked it (
Note: there's a rumor floating around that FXAA isn't working properly and an update is expected to fix it. I don't know enough to confirm this.). Now, if you know anything about Skyrim, there's no menu option to disable VSYNC, apparently because the physics get wonky at high frame rates. Nevertheless, below are a few screen prints for some anecdotal evidence of SLI with 680s with VSYNC on and then off.
VSYNC ON 
VSYNC OFF 
Here is what I want you to take away from this, with VSYNC ON (the recommended/preferred setting), the environment is gorgeous and fantastically smooth. In addition, the cards are hardly breaking anything that dares to resemble a sweat.
With VSYNC OFF, the physics immediately starting acting pretty crazy, but we can see why. The FPS results show the cards just cranking out a high amount of frames. In short, play with VSYNC ON, but know these cards have no problems taking care of business in Skyrim.
Metro 2033 This game is not only a NVIDIA TWIMTBP game, it incorporates PhysX technology, tessellation, and heavy volumetric lighting. This game always puts up a fight and this time around, it still gave the 680s a run for their money.
Knowing the 680s would be challenged, I went ahead and set the core clock to +100 offset and the memory to +250. I took several screenshots, but I'm posting the lowest, highest, and subjective average fps observations. Metro 2033 was set to the default Very High settings, PhysX on. This enables all the eye candy the game offers, except 4x MSAA. I didn't want to mess with that because the default was AAA. I also could have tested FXAA through the drivers, but I chose to keep it as simple as possible to base the results off of.
Lowest Observed FPS 
Average Observed FPS 
Highest Observed FPS 
As you can see in the first screen shot, this entails a lot of volumetric lighting, shadows, additional processing with the gas mask on, and tessellation on the mutants. Also, don't forget a portion of GPU2 is provisioned for PhysX calculations. This first scene is always low FPS. The second screen shot has a lot of characters to model, volumetric lighting challenges, and shadows. This is about the average I observed. The final screen shot is the highest I observed.
This should give a pretty good indication of what to expect. Overall, I'm pretty pleased with the results as anything above 60 fps for this game is a win with all of its engine complexities.
Just Cause 2 Benches I've played Just Cause 2 a lot and really think they did a great job with the open environment, ridiculous movie physics, and playground/field day type of gameplay. This game was fun on 480s, had some difficulty above stock clocks on 580s, but it appears 680s have no issues.
For this game, I set Precision X to offset the core by +100 and the memory by +250. The in-game settings were set to 1920x1080, everything high or very high and all eye candy turned on and VSYNC off. For anti-aliasing, I used 4x MSAA. I then ran the Desert Sunrise and Dark Tower benchmarks. Below are the results.
Desert Sunrise 
Dark Tower 
As you can see, the 680s just plow through these very nicely. Finally, here is an in-game screenshot with Precision X overlay so you can get a feel for the utilization, temps, etc.
Crysis 2 I was searching for a game other than Metro 2033 and BF3 that could possibly tax these cards hard, specifically where fully using the framebuffer was a potential scenario at 1920x1080. I found it. Crysis 2 doesn't really tax the video cards processing bandwidth but the Ultra settings, DX11 tessellation, and high-res graphics pack do the trick. As I've done in the past few demanding games, the 680s were set to +100 / +250 core and memory clocks, respectively.
Here's the opening scene on the sub awaiting orders.
Right away you notice the framebuffer is 1491 MB, almost to the 1536 MB limit of the 580s. Utilization, temps, and everything else are right in line with what you'd expect, as well.
After moving through the first half of the map, you see it's climbed to 1908 MB of the 2048 MB framebuffer. Also notice, GPU Boost has the cards up and over 100% utilization with FPS around 130. I noticed through the FPS was anywhere between 100 and 180 depending on complexity but the average seemed to be 100-130 FPS. Again, pretty respectful for the demands of this game.
Finally, near the end of the map, we have the highest observed framebuffer at 1938 MB and 101 FPS. Not the volumetric lighting, tessellated objects, and water dynamics making this a fairly busy scene despite only 1 NPC behind the front of the bus. All in all, the 680s appear to chew up Crysis 2 pretty well and CryEngine 3 appears to utilize the cards to their potential, too.
Portal 2 / Left 4 Dead 2 (Source Engine) Valve's Source engine is getting a little long in the tooth though they keep updating it. So taking the two most recent Valve games, Portal 2 and Left for Dead 2, we get a pretty good idea what 680s can do in a slightly more dated environment. Conclusion: they get bored.
Less than 50% utilization, temps around 50*C, almost no GPU Boost, and 180 FPS tells the story. These cards are bored. Still, they render the environment great, don't they?
Second verse, same as the first. Except this time, clocks are dropped all the way to 897 MHz. They're practically falling asleep. Stock voltages, 50% utilization, banging out 220+ FPS... all day long.
<message edited by mwparrish on Saturday, May 26, 2012 8:24 AM>