All LGA2011 processors already have PCIe 3.0 built-in ready to use, except the nVidia drivers do not enable it with their graphics cards, thus you'd have to use a registry hack to re-enable it.  Just so you know, if you used an AMD card like HD7000's, PCIe 3.0 is default enabled in their drivers.  In otherwords, this is nVidia's fault considering that with every new beta driver, it still isn't enabled, i.e. no accident.   The nF200 chipset can only deal with 2.0 signals, otherwise there should be PLX chips or a revised "nF300", it can be changed in BIOS or via registry hacking.
 
There is a 3-page thread in the X79 series subforum that talks about re-enabling 3.0 for nVidia cards, I think you'll find the registry hack there but it isn't an easy process without patience.  No gaurantee it will work for SR-X though.
 
That said, the implication behind this review where even at 2560x1600 using a GTX690 for games, performance loss is less than 5% performance in PCIe 2.0 x16 versus PCIe 2.0 x16.  One could conclude you don't loose much with your GTX670 in a 2.0 x16 slot because it basically half the performance of GTX690.  Computational work like general CUDA or folding will loose a bit more though.  But it is up to you if the registry hack is worth the effort.

Iehpron: Thanks for your reply. Yes, I already discovered the X79 thread and the registry hack (which I have tried with partial success). My problem is that as soon as I enable Gen3 in the BIOS my system becomes unstable. I guess I was wondering if there was a reason nVidia is being so reluctant to enable Gen3 in their drivers. Is there a more fundamental problem with Gen3 support in the GTX 600 series and/or some Intel processors? If I switch from the Supermicro board to the SR-X then has anyone tested this configuration and found it to work reliably at Gen3 speeds? I realise that the difference when gaming is marginal but this is for a HPC application using CUDA. I need total reliability but if I can get an extra 10% performance then I'll take it.
 
CyberstormXIII: I see from your sig that you have an SR-X with two GTX 690s. Are they running in Gen3 mode without problems? Have you confirmed this with benchmarks?
 

CyberstormXIII: Thanks. So the EVGA GTX690 works on the SR-X. I've also heard reports of an GTX680 working on an HP ML350. I guess my problem is either my Supermicro motherboard or something weird about the EVGA GTX670 SC 4GB.

shawkie,
 
Did you have SLI working on that Supermicro board? Because I think they only support crossfire... not SLI.

shawkie

Iehpron: Thanks for your reply. Yes, I already discovered the X79 thread and the registry hack (which I have tried with partial success). My problem is that as soon as I enable Gen3 in the BIOS my system becomes unstable. I guess I was wondering if there was a reason nVidia is being so reluctant to enable Gen3 in their drivers. Is there a more fundamental problem with Gen3 support in the GTX 600 series and/or some Intel processors? If I switch from the Supermicro board to the SR-X then has anyone tested this configuration and found it to work reliably at Gen3 speeds? I realise that the difference when gaming is marginal but this is for a HPC application using CUDA. I need total reliability but if I can get an extra 10% performance then I'll take it.

CyberstormXIII: I see from your sig that you have an SR-X with two GTX 690s. Are they running in Gen3 mode without problems? Have you confirmed this with benchmarks?

I had similar stability problems when I moved my 4 GTX 670 4GB cards from my SR-2 to my x79 system and enabled PCI-E 3.0. It turns out that one of my cards would not operate at all in an x8 slot (and all of my other ones would). Not saying you do, but I had a bad card. The other three 670's are running like champs, but with that fourth card in I couldn't boot windows in PCI-E 3.0 and the system was very flaky.

pci 3.0, intel, nvidia 6xx' and x79 is a mess.
Might work, might not work, depends on card, mobo and cpu combination and specific instances
amd's 79xx cards have a much higher chance of working under this configuation.
 
There is a real problem with manufacturers releasing hardware that is out of spec or missing entire use cases.   The rush to be first to market seems to outweigh having drivers and hardware implementations that actually work.