I run mine at 100%SS (3172x3092 per eye) when offline against AI.
Track detail = High
Player detail = Full
Opponent detail = Low ,with texture override to medium ("Opponent Texture Override":1, in player.json)
Texture detail = High
Shadows = OFF (or Low)
Anisotropic Filtering x 16
MSAA x4
PP = Off
Everything else = Off
I can have 9 AI cars, 10 visible (to include my car, it has to be 10 to see all AI cars)
In start grid @ SPA pos 10, i get 10.6ms Gpu frametime (so thats very much on the limit) With the BMW M4 Class 1
I use Openvr FSR to sharpen the image, with 0.6 sharpening and 0.6 radius
Alternatively For multiplayer i can shave 1-1.5 ms off with having 80%SS and 1.1185 multiplier in FSR.
Why use >1 values in FSR you might ask?
Well i have tested FSR both ways.
The way it is supposed to be used originally with monitors, ie. use <1 for target (0.77 being the ultra preset), to render image at lower res than monitor, and then FSR upsamples it back to monitor res.
In this case being vr, back to 100%SS res in steam VR (or what ever you use in steam vr SS%)
BUT. This is when it gets interesting.. i get less aliasing with
using it in reverse.
So i set render target lower with steam vr (to 80%SS) and then upscale it back to 100%SS resolution by using >1 values in FSR.
For me this gives better image quality with less aliasing.
I can only guess why? But steam vr SS% seems to be much smarter at adjusting (lowering) the HMD's resolution, than FSR is.
FSR only sets lower render resolution target, but steam vr SS% slider seems to do more, as the picture in my previous message about different rgb channels for the barrel distortion correction suggests. This is just guessing about why, but the results are true non the less.
So how to calculate the correct >1 value
if you wanna try it this way?
FSR target (0.77 being the ultra preset) is multiplier to both x and y resolution (like PD in oculus is).
So just divide the
wanted end resolution in steam VR by
what you lower it to.
Either x or y is good as it is multiplier to both.
Example:
So i set steam vr to 80%SS and i want to get it upsampled back to 100%SS with FSR
100%SS =
3172x3092 per eye
80%SS =
2836x2768 per eye
3172 / 2836 = 1.1185
So i set FSR to 1.1185
That upsamples the 80%SS image back to 100%SS res (3172x3092)
For me at least there is noticeable difference in aliasing using it in reverse. Im on AMD, and i have heard nvidia user say the opposite, so you gonna have to test it your self as ymmv it seems.
I tested this in SPA right after turn 1, looking at the stands on the left, (on the downhill before
eau rouge), as the stands create most aliasing artefacts when using down->upsampling.
I'd say thats a good spot to do the test
Final note. I have noticed that many complain about the amount of aliasing FSR creates. That's is true, but mostly because the 0.77 is
wayyyy too much below 1 for VR as "ultra preset".
0.77 being multiplier for both x and y res, it ends up being as much as 59%SS out of 100%SS.
So i would use greater values than 0.77 to avoid huge amounts of aliasing artefacts.
0.8941 is equivalent to 20% decrease in render resolution in relation to 100%.
You get this by diving the res the other way, ie. 80%SS / 100%SS (2836 / 3172 = 0.8941).
