In my app, there's some scenario where I want to dynamically increase RenderScale to 1.5 to improve visual quality (on GearVR). This works perfectly for me. However, changing RenderScale back to 1.0 sometime causes the screen to flash (/w random image).
Have anyone seen this issue? I'm wondering whether this is a bug in RenderScale API or just simply unsupported scenario?
Anytime you change VRSettings.renderScale, it re-allocates the eye buffers, which can lead to dropped frames. But this ought to work without flashing a garbage image. It sounds like there may be a race condition between the re-allocation and the next buffer swap after it. We'll investigate. If you are varying the resolution on a frame-by-frame basis, you should use 5.4's VRSettings.renderViewportScale instead. It allocates only the largest necessary buffers, but uses a smaller viewport within them to save fill rate and bandwidth.
Thanks Dave, where can I find more info about the differences between renderScale and renderViewportScale? I noticed that renderViewportScale can be 0 to 1, what's the default value? (I assumed 1)
Per you note above, I could set renderScale to 2 (my max), then dynamically set renderViewportScale down to 0.5 in scene (or scenario) that I don't need the max level of sharpness. Is that correct?
where can I find more info about the differences between renderScale and renderViewportScale?
See https://docs.unity3d.com/540/Documentation/ScriptReference/VR.VRSettings-renderViewportScale.html and https://docs.unity3d.com/540/Documentation/ScriptReference/VR.VRSettings-renderScale.html.
what's the default value?
Yep, it's 1. That means the entire texture will be used for rendering. A renderViewportScale of 0.5 would use an 0.5x0.5 square inside the texture, resulting in a 75% pixel throughput savings. A renderScale of 0.5 would result in the same savings, but it would force a re-allocation of the eye buffers instead of using a smaller viewport within a large buffer.
I could set renderScale to 2 (my max), then dynamically set renderViewportScale down to 0.5 in scene (or scenario) that I don't need the max level of sharpness.
Yes, renderScale * renderViewportScale would cancel out to 1 in that case, so the sharpness would be the same as 1, 1, but there would be a larger memory footprint.
On more question, for the later scenario (renderScale:2, renderViewportScale:0.5), do I also pay performance penalty twice? (e.g. perf implication from setting non-default renderScale + perf implication from setting non-default renderViewportScale)
There are 2 different costs to consider here: 1) The number of pixels being rendered and texels being read. This is related to renderScale * renderViewportScale. So 2 * 0.5 = 1 should have no impact on this. 2) The memory footprint. This is controlled by renderScale and has nothing to do with renderViewportScale.
