I think I obsessed over the RAM question one day and found out that it’s diminishing returns past 3200.
Thank you for your kind words! I’ll add a note about this in the OP, it might help others too.
Yes, that’s my conclusion too. But my RAM is 2600 overclocked to 3200 resulting in increased timings (21 from 16). From my research I concluded the most I can hope for upgrading for native 3200-3600 RAM is 1% to 3% performance increase, and considering the cost of 64Gb RAM replacement I decided against it. But maybe if I decide to do a CPU upgrade, it’s worth to upgrade RAM as well, to compliment GPU upgrade and make sure RAM is not bottle-necking the system and CPU-GPU transfers. Normally I upgrade/rebuild my PC when I can get 40% performance increase (roughly every 2 years), as I work on it too every day (web design/graphic design/video production). Last update/rebuild was 19 months ago. But performance gain would not be as good as I usually go for. But MSFS is a good incentive if it gets me from slightly below 30FPS (horrible VR experience) to slightly above 30FPS (smooth reprojection) 
Mine is “Native” 3200, if setting the stock XMP profile means it counts as that.
Thanks for the tip. I might be making a contribution here to the Journal of Irreproducible Results but I found your suggestion helpful and I think an improvement over the settings I was using, 100/100.
I’ve only been through my settings once or twice so fingers crossed that everything holds up. But rendering at 70% TAA in the sim and 150% SS in OXR works if I turn the WMR Headset Display frame rate down from 90 Hz to 60 Hz. Otherwise, black at the side when I turn my head and everything stutters. I only have an RTX 2070S (8 Gb VRAM). i9-9900K(4.7 GHz), 32 Gb 2666 MHz RAM, 1 Tb non-system 970 EVO Plus ~devoted to SIM). 100 Gb rolling cache.
The 70/150 setting seems to work better if I run MSFS windowed for some reason in 2D mode (even though it’s not in VR mode!). Reprojection is off, too. Game Mode and limiting frame rate in Nvidia CPL don’t help nor does hardware-accelerated GPU scheduling.
I’m running terrain and object LOD at 100 and my settings are as previously described except I turned Ambient Occlusion and Contact Shadows OFF, set Texture Supersampling back to 2x2 and Texture Synthesis back to MEDIUM. Otherwise as here: New Nvidia Driver 461.40 - #62 by JALxml
The interesting thing about 70/150 at 90 Hz is when a frame freezes, the screen picture is REALLY sharp. So I wonder if the perceived problem in sharpness with the G2 is the lack of a smooth uniform transition between frames and that’s why for me that 70/150 at 60 Hz seems sharper than 100/100 at 60 Hz, because (ignorance may be showing here), 100/100 is trying to operate at full resolution but 75/150 takes advantage of post-processing upscaling and has a smoother transition between frames because it’s not so hung up computing a full-resolution frame?
None of this may hold up tomorrow when I try it again but I would say the overall scene, not just the sweet spot, definitely looks sharper at 70/150 than 100/100 (but not “crystal clear,” as you say for the 3090, sigh!). I was making comparisons flying the ICON A5 out in the country over mountainous terrain near Mann Gulch, MT, and the comparison for me is easiest to make in drone camera view behind the plane (END key for me). I think in the cockpit that the windshield glare, curvature, etc., makes it hard to judge the pure scenery view and going outside the plane is better for making comparisons. Comparisons were made by restarting a spawned-in-the-air .FLT flight after exiting and restarting the sim. For testing hardware accelerated GPU scheduling, I rebooted the computer after any changes.
2 Likes
In my opinion choosing TAA100 or not should only be based on EFIS, because like I’ve described, FS2020 is rendering EFIS at a resolution which depends on the FS2020 render scale, not the final view render scale (OXR SS or SteamVR SS):
In general (resolution):
-
For any EFIS aircraft, use TAA100 and adjust OXR/SS to keep your perf target.
-
For non-EFIS aircraft, take advantage of the CAS Shader and use TAA70 or even TAA60 and raise OXR/SS to the max you can for your perf target.
In general (motion smoothing):
-
With no motion smoothing visuals depends on TAA scale and TAA70 is the best perf/pixel ratio, but TAA100 is the sharpest EFIS.
-
With motion smoothing, perf depends also on OXR/SS scale (for up to 25% your frame time budget)
In addition:
-
TAA > 100 is increasing details but with additional rendering cost. It might be worth it depending on your hardware even if you lower OXR/SS down to 50% (and soon below this with the next OXR preview).
-
OXR/SS > 100 is marginally increasing details compared to TAA > 100 on the G2, but is increasing details more on the Index.
The above gives you the general guiding points you have to take in account to balance your perf/visuals. When testing all of them you’ll therefore find some which are as good as others per-wise but not as good visually.
With higher capable hardware than my test system, I’d recommend still starting with my base settings and strategically starting with the recommended resolutions I’ve published, and instead of raising resolution you might want raising any FS2020 graphics settings using mostly the pixel shader first (clouds, texture super-sampling - although 4x4 should be enough in most cases, contact shadows, bloom, reflections). Only when you’ve raised these you might want to raise resolution if perf permits.
Taking in account both what I could see in the headset and what I could measure with math, I’ve chosen to keep only 3 different resolution ratios in the end (3 for the G2 and 3 for the Index). But these were also chosen based on my test system hardware and depends also on the FS2020 Graphics settings chosen carefully to ensure having a minimum pilot-wise view while trading other rendering resources allowing raising the resolutions.
NB: I’ll revisit the Index settings because the latest SteamVR 1.16.6 runs much better and I can raise a lot the settings and the resolution now).
3 Likes
Thanks again @CptLucky8. This is pretty much where I’ve landed with my 5800X+3080. OXR 70, TAA100 with most settings med/high and a few ultra and NVidia CPL limited to 30fps.
With regards to motion smoothing, at these settings it’s more like a 40-50% hit suggesting it’s a function of not only resolution but also the ‘complexity’ of it. I guess that would make sense if it’s applying some artefact or position/depth level of interpolation. Having looked at the OXR overlay (ignored it before as I only use my right eye, but switched on the WMR mirror to take a look) it’s almost exlusively a GPU hit, rather than CPU. MSFS MainThread timiings are a little misleading in that they seem to include the motion smooting GPU time.
2 Likes
@CptLucky8. There is no question that you’ve identified some of the very best settings to use and helped us all understand a lot more of what’s going on. Even as you’ve lamented, though, there is a lot left unexplained by Asobo in the workings of MSFS VR.
I think one thing that gives food for thought is that you’re working with an RTX 2070 Super and the observations of some of the folks with an RTX 3090 or a 2080Ti are intriguing. Perhaps the resolution and clarity of a G2 might not be a straight out of the box hardware thing but somehow at least partly tied to available GPU and software settings for rendering.
If nothing else the observations of @samplitude14234 My 2070 SUPER VR settings and suggestion (Reverb G2 - WMR) - #417 by samplitude14234 and @Anarchae1 Reverb G2 Sweet Spot Discovery have me wistfully shopping for a 3090!
It still seems like there is still a bit to sort out - but that easy for an ignoramus like me to spout! 
(Ooh! that rhymes!)
1 Like
I’ve never tried Nvidia 457.30. I did an Nvidia “clean install” with no GeForce Experience. I compared 457.30 to a similar clean install of 461.40. My current experimental settings and rig (2070S) are as referenced just above in this thread, 70% TAA and 150% OXR but I turned Ambient Occlusion and Contact Shadows both on to LOW. Flying out in the country 1K to 2K off the ground at < 100 mph, 457.30 is definitely smoother but not as sharp as I find 461.40 nor the overall appearance of the scenery as striking, IMHO. (461.40 seems to be brighter with more contrast as well as sharper???) If I turn off custom OXR scaling and let the software pick a setting, it undoubtedly goes for a lower OXR % rendering as the scenery gets blurrier for both 457.30 and 461.40 (see CptLucky8 post: PSA: Reverb G2 small sweet spots, observations and solutions - #150 by CptLucky8). YMMV but my main aim is scenery gawking in places I’ll never go IRL. So I’m mostly flying slow and straight and level. For example, if you start out at Machu Picchu and keep following the Urubamba River, you’ll eventually end up in Brazil at the mouth of the Amazon River. So for that sort of pursuit, I can get by with my settings as I continue to experiment and contemplate getting a better GPU for VR.
One thing I’ve noticed in donning and removing my G2 quite a bit is that it’s too easy to move the IPD slider accidentally. I 've taken to periodically checking the setting in the WMR Headset Display settings. HP ought to provide a protective cap or something to lock the setting in place.
1 Like
I applied some electrician’s tape to prevent accidental IPD slider movement.
2 Likes
As we know, you can’t change OXR custom scale „on the fly“. You need to exit VR and restart VR for changes to have effect. Now, what about setting it to Auto? Is it constantly changing the scale automatically or does it set a resolution based on system specs? Do we know how and what it does in the background? Regards!
If you like flying really low for the thrill of speed, set Terrain LOD very low. I set it to 50 or less.
Visually, it’s not going to make a difference – you are probably turning off some details somewhere you can’t see. But it’ll help you get a butter smooth experience, or save the resource for something else.
To be honest even at altitude the ,main thing you notice with Terrain LOD set low is the trees and you can extend the tree draw distance with one of the tree draw mods with a much lower FPS hit than using LOD.
1 Like
@CptLucky8 Do you have any observations on changes in VR performance after World Update 3?
Before the update, I could sit on runway 31L at JFK maintaining 30 FPS with repro off. OXR 70, TAA 100. I could do this in the A32NX which is, to me, the ultimate test. Sitting at Heathrow gave me 35-37 FPS. I limited my FPS to 30 in the Nvidia control panel and was happily flying.
Since the update, I’ve lost around 5-6 FPS on average. The exact same conditions now give me around 25-26 FPS at JFK and around 29-30 FPS at Heathrow.
My specs are 5800x, 3080. Very frustrating when I had it all dialled in so nicely.
1 Like
@LucioTheBird I can’t really tell for now because I’ve mostly flown over UK, which is a place I barely flew over before WU3 (too much flying over there in XP11 VR due to ORBX TE 
) So far I don’t find any loss of fps in VR with my settings and the Index when flying outside of the London area (everywhere north of Liverpool was fine so far for me).
However there is a bug which is: all of a sudden the motion smoothing gets a hard time keeping up and moving the head is like seeing a dance floor with a stroboscopic light. To cure this I used to open SteamVR settings and disable motion smoothing but it was not always working, nor setting motion smoothing again.
Yesterday when this happened I quickly started fpsVR and it gave the answer right away: FS2020 was dropping every other frame like with the beta. During the beta it required starting Steam Client prior running FS2020 otherwise you’d have the same problem.
But I’ve found a way to solve it, working twice in a row: just press the WIN key to open the Windows Start menu. In doing so FS2020 window looses focus, and immediately cures the dropped frames. Press again and repeat until is cures itself.
I noticed the same issue, and discovered the same cure, even before WU3. And I am using WMR OpenXR.
I thought I was dealing with the mouse bug you discovered, and talked about this in a previous post.
1 Like
My guess would be that it’s simple: it’s not a bug - London POIs and photogrammetry is much more complicated than before, of course it costs you a few FPS when all that is loaded.
Not really when you’re sitting on runway 27L at Heathrow facing away from all the photogrammetry and far enough away from it that it’s not even loaded in. Heathrow is way outside the city centre. And as I said, it’s the same at JFK. I’m down an average of 5-6 FPS across the board in all locations in VR.
I think it’s still very likely that some of the enhanced scenery is loading. Even the enhanced terrain mesh can cost FPS, and there are many things that can load at a distance. JFK for sure has Manhattan on the horizon affecting FPS. Not visible or obstructed by buildings doesn’t mean it’s not loaded or not eating FPS. I mean, the update that enhanced UK, put photogrammetry to London and introduced a lot of POIs that lead to people complaining that flying around London tanks FPS very likely has something to do with what you are describing… Much more likely than some new performance issue. Of course, generally the performance under VR is not good and needs to be improved. But this dows not sound surprising…
I know its not directly correlated to VR, but it seems like every update i lose 3-5 fps on the 2D test that i have, though there could be a connection with some hiccups i experience now in VR. (my 2d test is m20R at eglc, which now is 48 fps at 4k with render scale at 90 in 2d, most things on high, vector data at ultra, though with orbx eglc in play its around 45 fps, at one point bare eglc this was at 57 fps on 3080 ftw3)
Cruising around London with render scale 90 and terrain i think around 100, Openxr now at 50% (bug fixed) and everything on auto, i can fly in bonanza from eglc toward the west over the city with a scattering of juddering here and there, not horrid, smooth at times then a hiccup or two. I didnt perform the same test before the update though, so unsure if its worse/better in VR. I will try the Win key trick to see if it smooths things out.