Well there's the performance and then there's the experience. With Kepler, you're likely to vomit all over the place after extensive VR use (24ms latency higher than Maxwell). Maxwell is a little bit better in terms of motion sickness due to a better support of Asynchronous Time Warp using coarse grained preemption.
According to Nvidia, its VR-optimized Maxwell GPUs can reduce latency by as much as 24ms compared to earlier GPU generations. This is a major deal when you figure that good VR experiences are aiming for 20ms or less of latency, and current Nvidia GPUs are part of a pipeline that is 57ms total.
That's as low as 34ms for Maxwell. VR aims for 20ms.
GCN leaves both in the dust due to the finer-grained preemption used for Asynchronous Warp (yes the ACEs). Heck LiquidVR won a prestigious award:
http://www.roadtovr.com/amds-liquid-vr-initiative-wins-ais-lumiere-award/
This is why AMD are pushing VR, their architecture is better suited for it. LiquidVR is the better VR implementation when compared to Gameworks VR.
http://www.anandtech.com/show/9043/...ed-amd-gets-expanded-vr-headset-functionality
In the context of virtual reality, when rendering images to virtual reality headsets, low latency is critical to avoid nausea and motion sickness. One of the challenges associated with reducing latency is ensuring that the GPU has access to the latest head position information when it is rendering each frame. Even if the information is fresh when rendering begins, it may already be stale by the time the frame completes.
Asynchronous Time Warp addresses this issue by obtaining fresh head tracking information after each frame finishes rendering, using it to warp the frame to appear as if it was rendered from the new viewpoint. This warping makes use of a compute shader, and needs to be executed with high priority to avoid adding latency back into the pipeline. Executing it asynchronously allows latency to be minimized and helps eliminate stuttering, since context switching and pre-emption overhead can be avoided.*
http://developer.amd.com/tools-and-sdks/graphics-development/liquidvr/
So while FPS is important for running the games in VR, latency is far more important when it comes to reducing nausea caused by motion sickness.
AMD have planned for VR way before NVIDIA. Don't take my word for it, Oculus said this:
http://www.tomshardware.com/news/oculus-oculus-connect-vr-amd-nvidia,27729.html
According to Tom Forsyth, Oculus VR's Software Architect, this is a new enhancement for Nvidia GPUs that is*already*available in AMD graphics cards. In fact, some of these optimizations can be found in current Xbox One and Sony PlayStation 4 platforms as part of their respective SDK. Why? Because they are based on AMD GPUs.
In my opinion, the Nvidia reveal is a good news story for everyone. It's less about saying one GPU brand is better than another for VR; instead, this is about introducing a new choice and widening the landscape for GPU consumers interested in VR.
So for NVIDIA, great VR support really hinges on Pascal. So eventhough Kepler, if overclocked, can perform nicely... The latency associated with a Time Warp makes it relatively useless for VR.
Maxwell should be descent. A large improvement over Kepler for sure. GCN, well, it should run VR quite well.
As for Polaris, better than GCN.