I think I posted about this almost a year ago, but there was never a solution.  So I recently got a pair a GTX 980 Ti's and I'm experiencing the same old issue I did with my GTX Titans, which is a significant performance drop in gaming and benchmarks when my ROG Swift is plugged in.  Note that this is irregardless of whether or not G-sync is turned on or off in the NVCP or if it's an inactive display.  It only happens when it's plugged in and thus the G-sync options are available in the NVCP.  It seems to be driver-related and it still hasn't been "fixed" for me.  I used to think maybe it was my old Titans, but it's happening with brand new 980 Ti's.  
 
One thing I noticed is that under GPU-Z and NVCP, the Bus Interface is PCI-E 2.0 instead of 3.0.  This prompted me to the following fix:
http://forums.evga.com/How-to-enable-PCIE-30-in-Windows-8-and-81-Regular-and-Pro-m2063059.aspx
 
However, there's still a performance hit of 5% (which is much better, but it's still there and it's really bugging me).  I did this a few days ago.  This morning, every game I started up was crashing at some point, so I figured maybe this "fix" has externalities of some sort.  Is there a better way to enable PCI-E 3.0?  Just to be sure it isn't something else, I performed a clean install of Windows 8.1.  The issue persists.  I'm also having stability issues, as the display drivers stopped working randomly then came back on, leaving my screen frozen for about 15 seconds.  Are the most recent Geforce drivers a mess or something?  For reference, here are my relevant system specs:
 
Display(s): Asus PG278Q & Asus PB278Q
GPU(s): 2x GTX 980 Ti 
Mobo: Asus Rampage IV Formula
CPU: i7 3930K @ 4.3GHz
Memory: Corsair 16GB @ 1866MHz
PSU: Corsair AX 1200i
 
Note that the EZ Plug above my top graphics card on the mobo is not plugged in.  I used to always plug it in, but it doesn't seem to make a different.  I don't think it causes a lack of power delivery, but correct me if I'm wrong.

bob16314
In the BIOS ->  Advanced -> System Agent Configuration settings, set PCIE_X16 Link Speeds to [Gen 3] if not already set.
 
Instead of manually regediting, grab the force-enable-gen3.exe near the bottom of the webpage and run it.
 
Uninstall the Graphics Driver in Safe Mode using Display Driver Uninstaller (DDU) and then perform a Custom (Advanced) install of Hotfix driver 353.38, selecting only to install PhysX (unless any of the other optional driver components are needed) along with the Graphics Driver..Then run force-enable-gen3.exe.
 
Then run the Render Test in GPU-Z and see how it rolls.
 
Just a few things to try.

Thanks for your reply.  The force-enable-gen3.exe didn't do anything.  I restarted my PC after, but still no change.  I downloaded it, then opened it.  I tried it multiple times.  Perhaps it's outdated?  Please note I'm using Windows 8.1.
 
Update: Got it to work.  I had to run it as an admin for it to work.  I'm going to run 3DMark and see if this changed anything.
 
Update: The G-sync display hit is still about 5 percent.  Does anyone else have this issue or can they try it out?  Note that this issue only persists in SLI.
 
(1) Run 3DMark with G-sync off, but your G-sync display plugged in (does not have to be active, just plugged in)
(2) Run 3DMark with G-sync display plugged in.  Make sure settings and overclock are exactly the same as in the first test.
(3) Compare the two graphics scores.
 
Like I said, it's been almost a year and changing the graphics cards didn't change the problem.  PCI-E 3.0 helped, but still didn't eliminate the disparity.

Keep in mind that when you run the "force-enable-gen3" patch, you need to run it "as administrator" for it to work and then reboot.

bkhan530
I'll test this in a few as I have a non-gsync display right next to my PG278Q and those exact models of the 980 in SLI.
 
Edit: Here are my results with FireStrike:
 
Drivers: 347.09 WHQL
System: Intel 3770K @ 4.5 GHz, 16 GB 2133 MHz ram, Asus Maximus V Motherboard, 2 x EVGA GTX 980 SC with ACX 2.0, 
Monitors: LG 27EA83-D (IPS), ASUS ROG PG278Q (G-Sync)
 
ALL RUNS DONE USING STOCK CLOCKS WITH SLI ENABLED
 
Run 1: Multimonitor setup with LG as slave and ROG as primary: http://www.3dmark.com/3dm/5565202
Run 2: Multimonitor setup with LG as primary and ROG as slave: http://www.3dmark.com/3dm/5565246
Run 3: LG connected to GPU, ROG physically disconnected from GPU: http://www.3dmark.com/3dm/5565273
Run 4: ROG connected to GPU, LG physically disconnected: http://www.3dmark.com/3dm/5565303
 
Compare link of LG only vs ROG only: http://www.3dmark.com/compare/fs/3829337/fs/3829356
 
While there is certainly a hit with G-Sync around 4-5% (graphics score), I don't see the purported 30% hit that this guy is talking about. Regardless, it seems clear the G-Sync module introduces overhead and NVIDIA should focus on optimizing that, especially with FreeSync coming.

 
Can't believe they haven't fixed this yet. bkhan530 reported this in January of this year.

gridironcpj

Sajin

bkhan530
I'll test this in a few as I have a non-gsync display right next to my PG278Q and those exact models of the 980 in SLI.
 
Edit: Here are my results with FireStrike:
 
Drivers: 347.09 WHQL
System: Intel 3770K @ 4.5 GHz, 16 GB 2133 MHz ram, Asus Maximus V Motherboard, 2 x EVGA GTX 980 SC with ACX 2.0, 
Monitors: LG 27EA83-D (IPS), ASUS ROG PG278Q (G-Sync)
 
ALL RUNS DONE USING STOCK CLOCKS WITH SLI ENABLED
 
Run 1: Multimonitor setup with LG as slave and ROG as primary: http://www.3dmark.com/3dm/5565202
Run 2: Multimonitor setup with LG as primary and ROG as slave: http://www.3dmark.com/3dm/5565246
Run 3: LG connected to GPU, ROG physically disconnected from GPU: http://www.3dmark.com/3dm/5565273
Run 4: ROG connected to GPU, LG physically disconnected: http://www.3dmark.com/3dm/5565303
 
Compare link of LG only vs ROG only: http://www.3dmark.com/compare/fs/3829337/fs/3829356
 
While there is certainly a hit with G-Sync around 4-5% (graphics score), I don't see the purported 30% hit that this guy is talking about. Regardless, it seems clear the G-Sync module introduces overhead and NVIDIA should focus on optimizing that, especially with FreeSync coming.

 
Can't believe they haven't fixed this yet. bkhan530 reported this in January of this year.

I doubt it'll be fixed anytime soon (if at all).  We're such a small subset of consumers, so Nvidia could probably care less if their top 1% are getting 5% worse performance with g-sync enabled. 
 
Here's an interesting thought though: Do you think benchmarks in all of the 980 Ti reviews are hampered by this?  Most reviewers have g-sync monitors and I'm sure there use them for their tests (but with g-sync off), which would lower the 980Ti performance by 4-6%.  I haven't seen any overclock results that match mine, so perhaps that is the case.  I think it's something to consider.

Definitely something to consider. 980 Ti & Titan X are really pwning the Fury X then.

gridironcpj

Sajin

gridironcpj

Sajin

bkhan530
I'll test this in a few as I have a non-gsync display right next to my PG278Q and those exact models of the 980 in SLI.
 
Edit: Here are my results with FireStrike:
 
Drivers: 347.09 WHQL
System: Intel 3770K @ 4.5 GHz, 16 GB 2133 MHz ram, Asus Maximus V Motherboard, 2 x EVGA GTX 980 SC with ACX 2.0, 
Monitors: LG 27EA83-D (IPS), ASUS ROG PG278Q (G-Sync)
 
ALL RUNS DONE USING STOCK CLOCKS WITH SLI ENABLED
 
Run 1: Multimonitor setup with LG as slave and ROG as primary: http://www.3dmark.com/3dm/5565202
Run 2: Multimonitor setup with LG as primary and ROG as slave: http://www.3dmark.com/3dm/5565246
Run 3: LG connected to GPU, ROG physically disconnected from GPU: http://www.3dmark.com/3dm/5565273
Run 4: ROG connected to GPU, LG physically disconnected: http://www.3dmark.com/3dm/5565303
 
Compare link of LG only vs ROG only: http://www.3dmark.com/compare/fs/3829337/fs/3829356
 
While there is certainly a hit with G-Sync around 4-5% (graphics score), I don't see the purported 30% hit that this guy is talking about. Regardless, it seems clear the G-Sync module introduces overhead and NVIDIA should focus on optimizing that, especially with FreeSync coming.

 
Can't believe they haven't fixed this yet. bkhan530 reported this in January of this year.

I doubt it'll be fixed anytime soon (if at all).  We're such a small subset of consumers, so Nvidia could probably care less if their top 1% are getting 5% worse performance with g-sync enabled. 
 
Here's an interesting thought though: Do you think benchmarks in all of the 980 Ti reviews are hampered by this?  Most reviewers have g-sync monitors and I'm sure there use them for their tests (but with g-sync off), which would lower the 980Ti performance by 4-6%.  I haven't seen any overclock results that match mine, so perhaps that is the case.  I think it's something to consider.

Definitely something to consider. 980 Ti & Titan X are really pwning the Fury X then.

 
Report driver issues directly to NVIDIA: http://www.nvidia.com/obj...rqualityassurance.html
It makes sense honestly.  Nvidia, fix this please.  Your card will look even better in comparisons.

I get higher 3d mark scores by 5% if I unplug all my monitors and just run 1. Even higher if I use HDMI over display port. Weird.

I figured I'd revive this thread with an update.  I recently purchased an Asus PG348Q (3440x1440 100Hz g-sync display), and now my performance drop due to g-sync is 10-15% lower under the same clock speeds.  Note that I'm still running the same 3Dmark Firestrike test as before.  Would anyone else with a 3440x1440 g-sync display be willing to replicate these tests?  
 
(1) Run 3DMark with G-sync off, but your G-sync display plugged in (does not have to be active, just plugged in)
(2) Run 3DMark with G-sync display plugged in.  Make sure settings and overclock are exactly the same as in the first test.
(3) Compare the two graphics scores.
 
Taking note of Sajin's response, it's possible that Sandy Bridge E is the culprit and doesn't truly use PCI-E 3.0 since his performance hit is within the margin or error.  If so, it seems like the g-sync SLI performance hit is higher when on a higher resolution and using Sandy Bridge E.  

Sajin
Hmm. Not good.

When you replicated the test, were you using a 5960X?  

gridironcpj

Sajin
Hmm. Not good.

When you replicated the test, were you using a 5960X?  

4790k.