@stekusteku
This is a very interesting subject matter, further compounded by different factors and where theory and practice sometimes are telling different stories. Here are a few elements behind my reasoning.
First this is a recommendation not just when using motion smoothing, although in this case it is helping because motion smoothing computation depends on the OXR resolution, because it is computing motion vectors from the final output image. This means the smaller the OXR resolution, the better for motion smoothing. Of course a 30XX will do better than the 2070S here and will give more leeway in term of resolution.
Another factor not to forget about is the āqualityā of the FS2020 rendering engine. Iām not talking PBR or nice clouds here, but signal to noise ratio per pixel instead - or something along these lines. In other words, how much of each pixel is actually representing the entirety of the visual signal underlying that pixel. For example, if you have two vertical lines at a distance and there is only 1 pixel covering the two lines (mathematical ideal lines), the only way to faithfully representing the entirety of the signal represented by this pixel is to average all the contributors (the 2 lines). There are theories behind signal sampling thanks to Nyquist which tells you need at least 2x the maximum linear signal frequency (4x therefore for bilinear sampling - X and Y in a plane). As a matter of fact, super sampling 200% is a minimum for any renderer but it is expensive, and this is why there is a ton of techniques to make quasi similar looking renders with much less computations
Finally letās not forget one factor for such recommendation is the G2 itself. The only reason OXR100 is about 1.5x the actual panel res (2160x2160) is to compensate for lens distortion. In short, you take the flat projected image, then you have to spherically distort it in an opposite direction than the lenses distortion, so that when viewed through the lens, both distortions cancel out. In other words, you have to over sample the rendering if you donāt want to loose details at the center. But here is another factor that kicks in with the G2: its narrow sweet spot and narrow circle of details. In practice if youāre displaying the image at 1.5x res to compensate the lens distortion, youāll most likely observe a small highly detailed circle in the center and the surroundings will be perceived as blurry by contrast.
So what does all this has anything to do with the question?
In short here is what I think is happening when using TAA150+OXR50 (given the above):
- youāre sampling each pixel with more data and therefore youāre getting more details per pixel
- in addition with OXR50 youāre still rendering as many pixels as the native panel resolution. In practice it looks like youāre still getting more information per pixel
- super sampling TAA150 is also averaging out the tree leaves and all the transparent edges better. In FS2020 they cut corners for speed and there is no real alpha blending but cookie cut leaves instead.
- finally, the small circle of details is less apparent and the image looks more seamless on a wider area outside the small circle.
So in practice, I donāt find youāre trading much quality due to the lower resolution because it helps lowering the G2 sweet spot flaws, while at the same time giving better signal sampling and while also being comfortable to enable motion smoothing on the 2070S.
But beyond the theory and my own personal interpretation of the optical and digitized signals going through the chain, it is just best to try it out and compare how you feel it is.
Having said this make no mistake: it is helping getting an overall better VR experience with the G2 and the 2070S to me, under a certain number of constraints and goals. If youāre hardware allows you might still want to compare then TAA200+OXR50 vs TAA100+OXR100. You might be surprised!
In any case the general principles from my initial posts remain: you need TAA100 at a minimum for the best EFIS legibility. In this case youād adjust OXR res. to compensate within your hardware limits. You can still find TAA70+OXR70 pleasant enough even with EFIS for example, but you canāt beat TAA10 there. After a while with the Index and the G2 it is still remarkably better with TAA100. Maybe once they really take seriously the idea to implement FSR at least (1 day work according to most devs) and DLSS (2 days work now with the latest SDK), weāll have other means to compare, but although the Asobo TAAU is quite good in itself in 3D, it is lacking in VR.
I think that what could be valuable in our collective struggle with VR at this stage, is that we can compare apples with apples. In other words, the main difference between our systems is mostly the CPU+GPU combo and this is what affects the most the end result. I believe if all of you reading this could start with the same settings as a basis, just to have a common reference to build upon, and then try to see up to which point you could adjust/balance TAA and OXR resolution with and without motion smoothing, this should give serious trends and hints about what we could adjust further and why.
Iām not saying my recommendations are the best in absolute though, but Iām putting enough reasoning and testing behind to validate some choices, not just by their hypothetical book value but by their actual technical and physical implications in practice. I might be wrong in many of my assumptions as well but given the feedback on āMy VR Topicsā I believe Iām on good tracks overall. The settings Iām recommending above are showing extremely good similar results between the G2 and the Index and this further tells me they are hitting the sweet spot for the 2070S in terms of GPU use with SU5. Since most are also already set to High there is not much need to try building a settings combo with ULTRA in mind either. High is enough and Iāve been even surprised to realize texture synthesis is as good in High or Med even with the G2 resolution. Therefore from these settings, it appears the most valuable opportunities are about resolution and motion smoothing, and this most likely will only be video card dependent, and what we can try collectively to finding out if we experiment with the same settings as the same basis!
PS: since the time Iāve posted this, Iāve made further progress with this question and I can now run with motion smoothing enabled with OXR100 and adjusting the rest