So to test what kind of performance I would get on a high res oculus I set the res to 2560x1440 with everything maxed and oculus mode, I have a DK1 connected and displayed it on my main screen, I get about 4 fps in galaxy map and around 10-20 or so in the cockpit. If I turn off oculus mode I get around 90 fps
So any ideas what the main bottleneck is? Oculus drivers? I guess the SLI mode in VR doesn't really help much from what I have heard.
My guess would be a number of factors
Resolution:
2560x1440 is something like 1.7 times the number of pixels of 1920x1080 so that is many more pixels to fill per frame, not to mention if you are using SSAA or MSAA the number of pixels rendered/calculated jumps even higher.
Framerate:
Although it doesn't apply in your case as you aren't able to maintain a high framerate, VR is generally more demanding as it requires a higher Hz of say 75 fps for smooth motion. Considering most games only target 60 fps and many console 30 fps its quite clear to me that modern games are pushing the technology so much that framerate generally plays second fiddle to visuals. This will be a problem for future VR especially at larger resolutions or even higher framerates as you are having to do the same work in a much shorter time.
Two cameras / view points
This is an area where I suspect some optimisations could take place as I feel the generalised 3d engine will evaluate each view/eye separately even though 95% or more of the data for one eye is relevant to the other. In a generalised engine this would not be an issue because using multiple cameras normally means they would be very diverse views of the scene and thus would share little in common. So what might be happening is for each eye scene culling is performed, the per eye shadow mapping, per eye particles/billboards, per eye transparency sorting etc, you get the idea, essentially twice the work. My suspicion is that much of this work could and should be shared between views, though i'm unsure what level of performance gain if any it might provide as there are simply too many factors to account for.
Shadow Mapping
I really need to do some proper testing, but in general it seems to me that shadow mapping is one of the bigger factors in performance. This is for several reasons, for one the higher the quality of shadow map the more pixels it must calculate, which I think will be multiplied by the number of shadow casting lights, so that can add up quickly. You then have the issue of whether the engine is optimised to account for the fact that its displaying two different viewpoints as I described above.
Fillrate
An area I suspect ED suffers from more than other titles is one of fillrate due to the fact that much of it takes place in space or highly concentrated/dense models (e.g. spaceports). The problem with fillrate is that as resolution increases performance drops and I don't believe it is necessarily a linear issue. Overdraw can also be a bigger issue where there are multiple transparent billboards or full screen quads being rendered.
These things are not affected by hardware or drivers so they are always going to be an issue.
As to your specific issue, its unclear if by Oculus Mode you mean Direct mode, which is known to currently have issues and very poor performance, extended mde is the best one to test with. I'm also usure how you are display the rift view on screen and whether that itself might be adversely affecting your performance. SLI should be working quite well in ED, so i'm kind of surprised at your framerate being so low. In the near future nvidia will push out its new VR Direct mode which should in theory make SLI GPU for VR far better in terms of performance and lower latency.
What would be interesting is to perform the same tests but lower your resolution, to 1920x1080 and see how that compares, then try another test setting shadows to low and/or off, disable ED anti-aliasing, drop Material qulaity from ultra to high or medium etc. That should help to give you an idea of where the issue might lie. I'd also be tempted to disable SLI just to rule that out as an issue.
