Here is a quick update from my tests without motion smoothing. It is also preliminary and not entirely conclusive, but there are a few things important I can notice that I’d like to share.
NB: all these tests were conducted with the FBW A320 at different locations and airports I have a habit to fly to for the tests. If you’re flying a C152 steam gauges over a desert, you should be able to push things up.
With SU5 and OXR100, I find TAA85 a minimum to get nearly similar visuals as with TAA100.
With the same settings I’ve published a few posts above, but without motion smoothing, I can manage to tolerate TAA80+OXR100 up in the air but with my test hardware it really is starting to get more comfortable at TAA75+OXR100 overall except when maneuvering on the ground
It takes TAA65+OXR100 to start having enough fps to make it tolerable on the ground, especially when turning. However at these resolutions I get lower visual details than TAA100+OXR50. The latter also allows me to enable motion smoothing.
My conclusion: I get better visuals and VR experience with TAA100+OXR50 because this allows me to enable motion smoothing, while preserving EFIS legibility and with a good detail per pixel ratio. Otherwise it takes too much TAAnn when using OXR100 just to get similar or better visuals per-pixel. And in the case of OXR100 I can’t enable motion smoothing and when balancing juddering and motion smoothing artifacts, I prefer the latter in general (and they are mitigated on the 2070S with TAA100/OXR50 ratio and my settings).
After all these tests without motion smoothing I’ve compared once again TAA150+OXR50, TAA125+OXR50 and TAA100+OXR100. The latter has a little bit more details in the centre but it is having a shorter center circle of details falloff. The former is really close but with a more gentle falloff to me. There might be other caveats though (see post above) therefore it is really just for experimenting for now.
Last but not least, I was asked earlier whether I’m experiencing any shimmering and aliasing and I’ve answered no. When trying these TAA < 100 with the G2, I’ve now seen them! In short it is very simple: the buildings are reflecting too much the sun light (very bright highlights) and when using for example TAA70 the upscale process can’t create much details between one single very bright pixel and the next. It is therefore enlarging the bright edge and unfortunately it is also doing this without much possibility to “smoothen out” the edge. This renders once upscaled as aliased/stair steps along the bright edge and this is causing shimmering from one frame to the other as it would. This also explains why I’ve never noticed any because I’m always running at TAA100 with both the G2 and the Index! This is another good reason for me to prefer TAA100+OXR50 with the G2 for now
Thank you once again for all your expertise and time that you kindly contribute. An interesting note about MR on and MR off. I have an i9-9900KF @ 3.60GHz with a 2070 super, and prior to SU5 I finally settled on TAA70 + OXR100 with MR on, and with many MSFS settings on high. I found this to be the best experience for me, but after SU5 I had worse performance with exactly the same settings, most likely becuase pre SU5 I wasn’t GPU limited. After some experimentation I eventually discovered that TAA70 + OXR100 with MR off was much better and even gave me room to raise virtually all of MSFS settings to high. In fact, these settings seem to work better for me at the moment than even TAA100 + OXR50. I’ll carry on experimenting with your recommendations but this has been my experience so far. I wonder if it’s just a placebo effect for me, or whether it actually adds to the discussion and is something anyone else has experienced.
That’s what I thought for a while, then I realised the forests looked great with Contact Shadows (High at least is mandatory for me) even with Ambient Occlusion Off.
But I agree that with both off / low, forests look like 2D cardboard sheets.
Captain, have you ever tried it the other way around? That is, setting OXR high and TAA low. For example, my conclusion on my old headset (Dell Visor) is that OXR 180% and TAA 80% gives the best result in visuals - sharpness with very solid performance (using reprojection though). I don’t know how that would translate on the G2. Maybe something along the lines of 150% OXR 70% TAA. Could you please try it and report back. My final resolution reported by the game is around 2300x2300 btw. Also running an i7 9700k and RTX 2070 (non super)
I’m not the captain, obviously, but I think the whole idea is that MR works on OXR pixels, so giving it so many pixels would kill MR and eat FPS. The idea is to have MR work on 50% of native resolution, then the game upscale it, supposedly it’s giving out the best performance, letting MR work smoothly. I haven’t tested it yet, but it makes sense. I normally run 70OXR/100TAA MR on my 3080/5900X but I can’t say I’m very happy with the performance. Will try these new recommendations and report…
This is exactly what I’m doing with the Index: TAA100+SS152 (with motion smoothing)
However the G2 OXR100 is about 3.3K x 3.3K pixels and this is a lot already for the 2070S. But the equivalent of what you’re doing would be using TAAnn + OXR100 with the G2 (with nn < 100), and in this case, I’m not sure the cost/benefit is much better than TAA100+OXR50 (with motion smoothing) in practice: My 2070 SUPER VR settings and suggestion (Reverb G2 - WMR) - #729 by CptLucky8
Actually when set to OXR50 with the G2, you’re rendering slightly above the native panel resolution! (G2 OXR100 = 1.5x native panel resolution)
The tradeoff somehow boils down to:
better per-pixel details from a signal/pixel perspective but loosing a lot of this information between frames because of juddering
lower per-pixel details from a signal/pixel perspective but keeping most of this information between frames thanks to motion smoothing.
Beyond the technical prowess one can achieve in pushing the technical aspects, in the end the overall VR experience matters a lot and not just the pixels. Prior SU5 with my test system I needed to let go motion smoothing and because of this one of my strategy was to compensate with the rendering in order to maintain juddering to minimum tolerable to me (30fps/90hz). Since SU5 and the latest WMR/OXR I can manage to enable motion smoothing and the strategy is now to tailor the rendering in order to stay under the threshold where you switch to juddering (no motion smoothing).
In addition, I’m finding the G2 with motion smoothing is also less prone to camera tracking issues. They are still there but they get smoothed out and are less annoying than without motion smoothing.
NB: all these preliminary tests with the G2 are for now done with the FBW 320 because of the EFIS displays and the system load with this aircraft. Obviously should I’d be trying a light aircraft with analogue gauges, I’m certain I could balance things out differently, but maybe not much in the end. When doing my tests prior my latest post yesterday (trying without motion smoothing), I’ve noticed that up in the sky aboard the FBW 320 over the desert, I was still able to enable motion smoothing with my latest settings suggestion and in using OXR100 and TAA between 70 and 75 depending on the view complexity (looking inside, outside, etc…)
Slightly off topic but it would be interesting to hear what is the furthest and/or smallest bit of efis text or cockpit button that you can currently resolve at default camera position with your favoured settings? I have only a quest 2 but driving it hard and would be great to hear a expert report on what the other headsets are currently achieving
I can’t tell what is the smallest at all, and a few cm seated position difference can make a huge difference*, but in general I make sure my settings allows me reading the smallest text in the center EFIS of the CRJ or the A320. These are good calibration points for me. Sometimes they are just at the limit between deciphering and reading though.
*with the A320, the 3D model seats are reclined, therefore my seating position is usually at the seat edge, aligning vertically so that the top edge of the PFD/MFD are displaying right under the glare shield. At this position the alignement ■■■■■ on the center post are nearly aligned too.
No. That’s what i’m trying to tell you. Setting OXR to 100 and then ingame TAA to 125% is not the same as setting OXR to 125 and TAA to 100. If i understand correctly, if you set the OXR to a higher level res (say, 120% oxr) and then in game TAA to 100 or less what is actually happening is the higher resolution is getting downsampled to the resolution you set ingame. If you set OXR to 50 and TAA to 100 it’s getting upscaled. The point however is that if you’re downscaling from a high OXR resolution it seems to me visually a lot less information is lost and what i’m seeing is a far sharper image with far less shimmering from poor AA (especially on buildings). And the menus and ingame texts are incredibly sharper to look at. What is most interesting is that it doesn’t seem to affect performance a lot. So on my headset if i set OXR to 100 and ingame to 100% i think i’m getting a resolution around 2350x2350. However if i set OXR to 180% (what i’m running now) and ingame TAA to around 75% i’m ALSO getting the final resolution around 2300x2300. Just the result is a far sharper picture. So it doesn’t matter if at 100% the g2 is 3300x3300x. Just set it to 120-130% OXR range and then just use significantly lower TAA (maybe 50%-ish). I expect your image should look better at not a lot of cost in performance. I may be completely wrong here and i hope anyone reads this takes my suggestion with a grain of salt but at least on my old wmr headset the results are definitelly as described. I’ve tested it a million times with a vast range of different oxr and taa resolutions. Best regards Let me know if you try my suggestion how it works out on the g2.
The problem with the G2 is that 100 OXR is already 3.3K x 3.3K pixels and with motion smoothing this is a lot to compute with. Therefore and because the G2 is using 1.5x more pixels than its panels resolution, and this is a lot for a 2070S, the strategy I’m suggesting is to configuring it in a different way from what I do with the Index, yet, while preserving enough details for the overall VR experience.
I applied all your settings to the letter earlier this week. I flew around for several days, performance was great but the resolution was less than i expected. Hey ho i thought, thats the price i pay, this has been established, quit messing around and fly.
Performance was so good i thought maybe i can up clouds to Ultra, no problem, maybe i can push TLOD to 200. No problem.
Several days later i hit escape after doing something in the menu and im back in the cockpit about to take off. Then my brain said wait a min, what was that “65” i just saw. None of my settings should say 65…
Low and behold i adjusted everything but TAA from 65 back to 100. THERES THE CLARITY I WAS EXPECTING… I’ve since had to bump clouds and TLOD back down of course but well worth it. Needless to say i felt a little silly. The last few days ive been flying thinking “Is this really the EFIS clarity the Cpt is happy with, because its pretty bad!”
I’m totally speculating here, and may be entirely incorrect, but I don’t think this is how it works. Think of WMR headset as a monitor of sorts.
MSFS is outputting TAA% of the OXR resolution (OXR%). OXR then shows it on the “monitor” (WMR/G2) and upsamples or downsamples it if it needs to, after applying a barrel distortion correction, which is why 100% OXR is higher than actual native resolution. That’s all. There is nothing really to “downsample” if OXR is high and TAA is low because the information is lost. MSFS just generating low-res image that is upsampled by OXR (same as your monitor would upsample lower resolution output). If TAA is high and OXR is low, then OXR is downsampling it to it’s resolution after applying distortion correction.
The catch here is that because MR works within OXR/WMR framework, if that resolution is lower - it takes less resources, so it’s smoother and has a higher FPS overhead.
Also, in theory, it should look much smoother and sharper if you take a high-res picture (TAA) and then downsample it (OXR), then if you take a low-res picture (TAA) and then upsample it (OXR). The distortion correction and scaling to actual HMD resolution is complicating it of course, but the usual rule of GIGO (garbage in - garbage out) should still apply. The better 2nd stage image should result in better final image.
I will still have to test it in practice with 150% TAA / 50% OXR on my 3080 and see what I get. Normally I run TAA100/OXR70 and that gave me best (but far from perfect) results. I never tried TAA>100.
All your post is spot on! Thank you for wording it out in different ways!!
I’m highlighting this part of your post only for something really specific in my settings approach above (TAA100+OXR50)
When using OXR50, you’re rendering at slightly higher res than the native panel resolution. When the WMR compositor is taking the renderer image (about 2.4 x 2.4 IIRC at OXR50), it will barrel distort it to 1.5x native panel resolution. You might think therefore center pixels will be blurrier and you’d be right. However, since the lenses are compensating the barrel distortion with a pincushion inversely proportional to the barrel distortion, and since these “blurrier” center pixels are in fact conflated only, they are not loosing much information when pincushion back through the view lenses. In essence, it is just making sure you’re rendering enough pixels per physical pixels and given the G2 pixel density (compared to other HMDs), you don’t notice any loss in the center as much as with others (like the Index for which 2.2x2.4 render res is a minimum for the 1.6x1.4 panel res).
Mind you, on the G2, OXR50 will always look less detailed than OXR100 of course. But the tradeoff here is choosing to loose spatial details (motion smoothing on but lower res) instead of temporal details (juddering but higher res).
In practice, and for the overall VR experience, loosing pixel details because of juddering is worse to me than loosing pixel details because of resolution in this case, especially since OXR50 is still slightly above native panel res and the G2 panel res is already high density.
As a matter of fact, with the settings I’m suggesting above (TAA100+OXR50) I still get more details than with the Index which I’m pushing to 152% with motion smoothing (IIRC about 2800 x 2600)
The explanation you provided sounds very sound. And Captain like he said has elaborated on the mechanics of it before. However, it’s still a bit funny to me that nobody wants to try what i’m saying It doesn’t really matter what the headset resolution is. However high you set the OXR resolution, you can simply adjust it lower (to the same ingame resolution as you normally use) by setting the TAA resolution lower. So for example if you would set OXR to 150%, just set the TAA to whatever you need to play at the same resolution that you do in the vice versa scenario. But i’m telling you, the result is undeniable. High OXR, lower TAA provides a better image at the same ingame reported resolution. And i can’t really tell there’s much difference in performance. Please just try it
You are right, but it depends! if you follow all my posts on this subject, you’ll find out I was even recommending TAA60 + SS220 for the Index for ANALOGUE gauges aircraft. Best visual overall outside and inside.
However, the problem is with EFIS gauges, because the TAAU upscaller Asobo is using is not good enough and it is conflating thin EFIS lines to the point you’re loosing legibility inside the cockpit.
It’s true. The world is rendered much sharper but the gauges not so much. And i’m having trouble compensating for this on my rtx 2070. Ironically i would need just a 10% higher TAA resolution (at oxr 180%) for the gauges to be completely sharp. I compensated for this by completely disabling the volumetric effects in the developer menu which gives an insane boost in fps but flying in nothing but completely clear weather (visually) is very sad. I need ASOBO to provide a 20% boost in performance from what it is now then my VR will be perfect even on my trusty old 2070. However, people with g2 and 3xxx cards have more than enough headroom to compensate without disabling the clouds. It would seem though the majority of people lately have been complaining about the horrible antialiasing in the world rendering (esp buildings). And my suggestion does provide and immense boost in sharpening (still not completely without shimmering, but still uncomparably better than with usual oxr/taa settings.
