[SOLVED] Variable Rate Shading

[davide445]

Looking a way to reduce a big image rendering GPU load, an obvious mode will be using eye tracking (to be integrated using HMD specific SDK) and VRS.

This is proprietary Nvidia but didn't find other suitable options (except possibly when the new AMD Navi GPU will be presented).

Didn't find anyone writing about in the forum, want to know if someone does ever try or get an idea about how complex will be.

I remember @silentexpressing his concern about the complexity in integrating VRWorks in Unigine, but this need to be a single component where there seems to be (in my limited knowledge) a clear integration path.

Edited January 27, 2020 by davide445

[silent]

Davide,

As I mentioned before, most of these VR optimization features (including VRS) is not easy to implement with deferred rendering and full set of post and scree-space effects. To be able to "easy" integrate this you need to have a simplified (more likely forward) rendering.

So, to sum up:

• Big engine internals re-writing is still required (simplified forward rendering). This leads to:
  • limited number of lights (forward)
  • increased polygons count with each light (forward)
  • no (or very limited) post processing and screen-space effects (including TAA)
  • decals, water, volumetric objects and clouds more likely will be unavailable
     
• Implementing VRS in current fully deferred approach will not give you any benefits at all. And there is a small possibility that some rendering artifacts may occur (broken lightning, for example).

Thanks!

[davide445]

Thanks for explaining so clearly. Was also thinking about using this to reduce needed bandwidth for the CloudVR idea, but can achieve the same result with post processing.