Reverb G2 Sweet Spot Discovery

The settings I mentioned is 80-OXR/100-TAA (now I understand why CptLucky8 invented this naming scheme, brilliant indeed). :slight_smile:
I can’t really compare 6800XT to any NVIDIA as I’ve been an AMD user for the last 10+ years. My settings are pretty much all ultra+175LOD (terrain and objects), V-cloud (medium), AO - low. I also turn off AI/live traffic. This nets me a very fluid and consistent 40-42fps pretty much everywhere except in big airports (KLAX - mid 20s-low30). My previous AMD was Radeon VII which would be pretty closer to 2070Super or 2080. 6800XT in raw rasterization is about 80-100% faster than Radeon VII. The increase in performance scales differently from game to game, and MSFS is far from optimized both for CPU and GPU (especially for VR), so things should improve in the days ahead. Having tried multiple settings and combinations, mid 30fps is really the grail performance for this sim in VR at the moment. No need for motion projection as the headmovement in G2 still produces silky 90Hz. G2 has headroom for at least 2 generations of GPU in my opinion. I think the next iteration of Navi or RTX along with ongoing optimization of the sim will get us to the native 90Hz. Or maybe not. :slight_smile:

4 Likes

I’m glad that you have better results with 3090. But I still believe the “sharp center - blurry rest” on G2 is an optical issue.

You mentioned you have OXR on Auto. So probably OXR lowered your “post” resolution massively on 2080Ti. When you installed 3090, such auto-lowering was probably just less massive, due to higher specs of 3090 (VRAM).
To have any conclusive results, I suggest to:

  • set “Custom Render Scale” in OXR Developer Tools,
  • enable developer mode in the sim and display FPS.
    This way you will be able to see the resolution rendered by the sim and then enhanced by post processing.

For example, on my 3070 (not as powerful as your 3090) I have following results:

OpenXR: Auto
TAA: 60%
Game resolution: 1488x1454
Post resolution: 2480x2428 —> dramatically low, perceived as barely acceptable in the center and total blur at the edges

OpenXR: 100%
TAA: 60%
Game resolution: 1900x1857
Post resolution: 3168x3096 → much higher in post, more sharp accors the entire FoV, but still more sharp in the center and quite blurry out of it

OpenXR: 100%
TAA: 100%
Game resolution: 3168x3096 → even higher in the sim, but outside the capatibilites of 3070 (10 FPS)
Post resolution: 3168x3096 → much sharper in the cockipt (glass screens) bu still much shaprer in the center vs. the edges.

What probably happened in your case:
Your resolution was probably massively reduced by the auto OXR setting with 2080.
When you installed 3090 the auto resolution went up to more reasonable setting.
So you now perceive more clarity and apparently the bigger swet spot, compared to the previous, dramatically reduced resolution by OpenXR autosetting.

To answer the question if anything else plays role here (like a bug in the OXR driver applying lower resolution outside of the center) or rather the optical issue, you should pick a fairly small detail in the cockpit and then place it in the center of your field of view. You should then rotate your head and place the same detail at the mid distance and finally at the far end of your field of view and confirm if it remains sharp.

In other words: You were probably initially limited (on 2080 and OXR auto) with very low resolution rendered, well below the optical capabilities or imperfections of the G2 lenses. When your resolution went up to a reasonable level, the question is if you are fully satisfied with end-to-end sharpnees across the entire/most FoV. I doubt it. This is the optical limitation of G2.

I never used Valve Index, but people like @CptLucky8 who have both G2 and Index say that Index has more or less the same sharpness across the entire FoV (while less sharp than G2 in the center), which for me as the G2 user is something I can only dream about.

1 Like

When you say you have the OpenXR custom render scale disabled, do you mean you have:

  1. not checked the box; or
  2. checked the box and moved the slider to ‘0’?

It is my belief (though I have no facts to back this up) that when the box is unchecked the render scale is automatically set at 100.

Just answered my own question. If I check the Custom Render Scale box and attempt to move the slider to ‘0’ it will only adjust to a minimum of 50. So, unchecking the box is disabling the feature. Sorry to bother you with the question.

@Beulah6126, it is brilliant, but only once you understand it. Until then it just looks like technical mumbo jumbo that only the cool kids understand, and you’re not one of the cool kids.

But yeah, 2x my 2070 Super would explain the ability to do the higher settings and still get the fps you’re getting.

I believe I’ve written about this already in this topic here:
PSA: Reverb G2 small sweet spots, observations and solutions - #150 by CptLucky8

I’m not sure how you can use an argument that CptLucky8 gets improved Index resolution VIA SOFTWARE to make the argument that the G2 is OPTICALLY LIMITED. Perhaps similarly even with what we think is a great GPU/CPU combination, the G2 is still limited by both the computer hardware/software combinations currently available to us. I only have a 2070S with an i9-9900K but when I go from 100% TAA in sim/100% SS OXR (full optical resolution) to 70% TAA/150% OXR, I get a much better image overall in the G2 (sorry I don’t have an Index to compare!). Same optics in G2 but in 2nd setting, better software settings calculating lower res virtual images for each eye, then post-processing automagically upscaling them to a noticeably sharper overall image. I have also noticed if I really push my luck and go to 100% TAA/150% OXR, which is 100% STUTTERVILLE, the images that momentarily freeze on my screen seem incredibly sharp compared to lower settings that give smoother motion without major stuttering. I don’t think getting a smooth frame rate with a 3090 necessarily means one is getting the sharpest possible image out of the headset hardware (going back to CptLucky8’s Index image improvement in Steam via a software change…). The other thing that I wonder about in comparing Valve “sharpness” to G2 supposed blurriness is that with the Index you have larger pixels distributed over a wider field of view (4.6 M pixels in ~130 deg FOV vs 9.3 M pixels in ~114 deg FOV). The G2 pixels are going to be smaller, being crammed into a narrower field of view. If you don’t have really good corrected vision, they’re going to look blurrier (try reading 12 point font vs 9 point font). If you have less than perfect vision, the clearest small spots are going to be in the center (because of the greater sharpness of foveal vision, that’s where the limited optics of VR headsets aims to provide the clearest vision and the headset optics don’t change when you move your eyes) and those small pixels are going to look blurrier off to the side, less so the sharper your vision, e.g. 20/15 corrected vs. 20/35 (perfectly acceptable to drive with up to 20/50 vision in Texas). So, yeah, everything looks sharper with an Index under Steam when you only have to resolve bigger pixels!

Edit_Update: Don’t know if it’s possible in MSFS but one thing that might be useful would be if the SIM could present the view for either eye that it’s calculated for VR in motion in a 2D image at the same resolution, i.e., essentially a buffer that captures an image frame that would be sent to your headset and allows you to view that frame at headset resolution in 2D. Right now if we switch out of VR, we get on our monitors the image that’s supposedly going to each headset but I don’t think it’s at full resolution (my monitor is only 1440p high and the vertical height of the image is ~1/2 the height of the screen). So I’m wondering if we could see at full res in 2D the images that are being sent to a single eye, whether that would help sort out questions of computer/hardware software computation of image vs. G2 optical quality. I guess one’s monitor screen pixels are a lot bigger than those you’d be viewing with the headset, too

1 Like

I believe you’re half right and half wrong. Please let me explain some key elements missing which will help understanding some of this.

First and foremost, the FOV you’re mentioning are diagonal FOV only. This is very important because the Index panels are rectangular and vertical, whereas the G2 are square. These headsets physically designed FOVs are measured and reported here:

HMD Geometry Database | Collected geometry data from some commercially available VR headsets.

But what makes the Index optics also better for edge to edge clarity is that these are dual-element lenses, and the panels are canted (by 5 deg):

The custom lenses built into the Valve Index Headset maximize field of view without sacrificing edge-to-edge clarity. High geometric stability allows you to look around the scene by moving your eyes (not just your head) with minimal shape distortion.

Instead of a mount perpendicular to the user’s eyes, the headset’s displays cant outward by 5 degrees. This improves outer FOV while balancing the inner FOV.

Headset - Valve Index® - Upgrade your experience - Valve Corporation

This difference is in fact massive to the experience:

  • The canted displays are contributing to making the distance between the external side of the image (right side of the right panel) and the headset enclosure smaller, and this was a revelation to me: with the G2 the enclosure looks wider to the side and is contributing to the “see through goggles” impression, whereas with the Index they are small enough to give enough peripheral presence and contributing to the “being immersed” impression. The vertical FOV is also much higher and contributing to this immersion as well.

  • The lenses are effectively giving you the most of each pixel nearly edge to edge. To me it is about 85% off center before it starts getting blurry but even there, it is less blurry at the edge than the G2 at 50% off center (about). I also notice the Anti-CA filter in WMR is not as effective as the combination of both the dual-element lenses material and the Anti-CA filter in the Index (I’m sure there is one too, it is the same lead developer Alex Vlachos behind the two).

So in turn, although you never get a pixel as sharp as the G2 in the center with the Index, you get overall more of each pixel across the entire field of view and this is visually much more immersive and much more comfortable in practice (let alone my own G2 optics problems I’ve described - awaiting a printed slimmer mask to compare prior doing the RMA).

And if you add the rest: sturdy build, eye relief adjustment (distance eye/lens), the convenient button on the headset which allows me popping up the SteamVR overlay and changing setting without having to touch a controller, with the addition of the razor sharp tracking which is never failing nor lagging nor glitching… It adds up to the overall experience and I really want the G2 to be my main sim driver but the more I continue using the Index, the more I’m really starting to wondering whether the G2 is really up to the task in the end despite the panel resolution.

3 Likes

Yes this is possible and “only” requires building a proxy DLL intercepting the OpenXR API calls. It is exactly what programs like ReShade are doing for the 3D API DLLs. I thought about coding one like this but I’ve never allocated any time because, well, there is always something else to do :slight_smile:

2 Likes

There is also this in-depth discussion of the Valve Index in relation to FOV. I’ll have to see if I can come up with something similar for the G2 as to the pro’s and con’s of it’s optics choice (like no eye relief!).

Field of View - Deep Dive - Valve Index - Upgrade your experience - Valve Corporation (valvesoftware.com)

Seems like a good run-down on FOV considerations (explains HAM!):
(also by Richard Musil, same guy running the HMD Geometry Database quoted by CptLucky8)

VR headset rendered FOV calculation - VR Docs (risa2000.github.io)

2 Likes

I wanted to further comment on this and I’ve taken a screenshot for another unrelated problem which gave me the idea to specifically compare the G2 and Index headsets with this aircraft cockpit as a reference.

I’ll use this screenshot for illustration only (just to remind the disk of clarity limits I can observe)*:

I’ve compared both in placing me in front of the attitude indicator and looking directly to its center point, with the Cessna logo on the yoke visible near the bottom of the view, then in moving only the eyes around.

Reverb G2:

  • I can see the attitude indicator gauge sharp overall.
  • The right most side of the airspeed indicator is showing slight blur, increasing as you look further to the left.
  • The left most part of the altitude indicator is showing the same level of blur increasing as you look further to the right.
  • The 3 gauges below (Turn Coord, Compass, VSI) are slightly more blurry and their lowest part is becoming sensibly less sharp.
  • The RPM indicator is heavily blurry, you can still decode the writing but not read it at a glance.

To keep it simple, trying to read anything beyond the middle of the 5 gauges surrounding the attitude indicator requires moving the head.

Valve Index:

  • There is no blur vertically from top to bottom. I can read clearly the Cessna logo and even see no blur on the features on the floor.
  • I can see no blur from the left most gauge (fuel QTY) to the left edge of the Garmin GPS.

The Index has a large edge-to-edge clarity where you can read easily anything within the 3 black grooves as clearly as in the center, and anything quite far beyond still, with minimal blur. You don’t feel the need to turn the head to read better.

Comparing effective resolution:

  • In the entire Index region I’m describing, the effective perceived resolution is higher than on the Reverb G2 image seen when reading the “100 knots” mark in the airspeed indicator, or the aircraft symbol in the center of the compass.
  • In the disk of clarity region of the G2 I’m describing, the effective perceived resolution is higher than on the Index center, there is no question there.

Comparing optics (simple test in SteamVR):

  • launch SteamVR Home
  • set Alyx 17 environement
  • navigate to the 2nd part of the environment (after passing through the porch on the left)
  • go near the wall onto which passes a big cable
  • turn around and look at the wall at the opposite side where there is a Russian poster/writings
  • turn your head left and right and see how the wall and poster are geometrically deforming
  • remove the front mask and hold the headset closer to your eyes and repeat.

There is a distance were there is no deformation at all (in addition to increased FOV and the perception of a wider sweet spot we all know already), and if you get closer to the point your eyes are nearly touching the lenses, the wall deformation will come back but this time reversed (like farther away it is barrel distortion and closer it is pincushion).

I don’t see any distortion whatsoever with the Index regardless of eye relief adjustment. It is possible SteamVR is also varying the pre-distortion to compensate for the eye relief distance changes too, but I just think the optics have a better designed sweet spot allowing such variability optically only.

Do you see this happening with yours too?


*this screenshot comes from an unrelated discussion:
LOD Problems - Distances revisited - #32 by CptLucky8

2 Likes

Put your foot down @CptLucky8 don’t get trampled on haha. :joy:

What use is the render scaling in MSFS if you ignore the AA. Nobody is going to turn AA off are they so knowing what the AA is - is crucial when conveying image quality settings to other users. Haha sorry to stir the pot :slight_smile:

1 Like

It depends. For example I’ve tried using high super sampling lately and in this case, you might find FXAA gives good visuals and better fps.

The idea I was trying is finding the resolution I could reach with the same overall perf as TAA 100, thanks to freeing resources using FXAA instead of TAA, or would there be a super sampling scale with no AA (or FXAA) giving the same visuals as TAA.

I could then write: FXAA300 + SS100* :smiley:

It just makes things simpler to describe in a short form factor anyhow.

*these are not actual figures, just an example.

2 Likes

Yeah having been inspired to try from a few posts I was also looking at FXAA with higher resolution but for me it doesn’t cover up the artifacts as well as TAA so making that clear on your analysis posts was a good call. :+1:

1 Like

As amateur photographer and birdwatcher I can only confirm: when it comes to the optical equipment/lenses usually you get what you paid for. This probably applies to the optical quality of the lenses of the more expensive Index vs. cheaper G2.

1 Like

Most of the benchmarks I’ve seen put the 6800xt at slightly higher performance in frame rates than the non-OC’d 3080. My RX 6800 seems to sit midway between the 3080 and 3070. This is an averaging of FPS performance over a couple of dozen demanding games, including MSFS2020.

I also have a 10850K, but paired with an XFX Speedster Merc319 RX 6800, non xt card. My video card can OC, but not as high as yours. Do you think the OC’ing of your CPU/GPU contribute significantly to your results, or more due to your settings?

My OC on 6800XT puts the performance at 6900XT stock which translates to about 10% over stock. For the current state of the sim, every frame helps. CPU OC will not see a huge gain until DX12 implementation. Only 1 core gets close to 90% and the rest of the 3-4 cores hover around 25-40% usage. If you can push your 6800 to 2500Mhz, you should be able to reach 30-35fps.

One caveat: I am experiencing CTD in VR every flight. Not sure if this is due to AMD driver or the Sim itself. Anyone experiencing the same?

2190Mhz is the ceiling boost clock for my card. Not sure what my FPS is, but on most settings I’ve tried it seems pretty smooth with only a few settings producing stuttering or jittery motion. So far few of my flights have ended with me just leaving the game w/o a CTD, the most recent being a sunset soar around Neuschwanstein Castle in Bavaria in a Citation jet. Flew around that region for quite a while trying to find a runway and attempt landing. Maybe because it was night time?

BTW, AMD just released a new driver: 21.2.3 on 2/17

I am on 21.2.3. They have not fixed the issue.
You should be able to overclock manually to at least 2400Mhz. 2190Mhz is probably what the Radeon software is telling you as boost frequency, correct?