I noticed this frequent red/blue back and forth as well. I’m guessing this is an indication that something or other is at its limit and therefore a good thing (compared to being below the limit.)
While the switching was going on I did not notice any issues with the VR quality. I was tweaking my settings and settled on TAA = 100 and SS = 80. This is with a 3090 and i9 10900K.
Unfortunately my new Reverb G2 died a few days after I started using it. Bad 6 Meter cable. Big problem affecting many users. My cable is on backorder so I won’t be able to continue trying VR for probably a month or so.
I do want to thank you and the others who have posted about the settings. It has been immensely valuable. I was trying to obtain a video card for many weeks, after building my new rig, and was unable to do anything except read and study the posts. When I was finally able to find a card and use it, I had a very trouble-free experience due to applying all these optimizations from the start.
I think there also may be a bug somewhere. I have had times where my indicator will be blue for most my flight, then will start flashing blue/red. I can even restart the flight and it will stay flashing blue/red. The only way to fix it is to either exit VR and re-enter it, or sometimes it requires me to restart MSFS.
I’ve had my first dive into VR over the last couple of days and I’m absolutely hooked but also extremely disappointed in some ways. I’m on a 5800x, 3080 and Reverb G2. In 2D, I can run a mix of mostly ultra/high and have a smooth, stutter-free experience locked at a consistent 30fps. For the first time playing a flight sim, I’ve been able to just enjoy it and fly rather than constantly tweaking as the sim looks so good out of the box.
With VR…I had my expectations well in check, I knew I’d have to dial back to a mix of mostly medium settings with one or two on low or high, I just wanted to maintain 30fps and let reprojection do it’s thing. My setup is optimised in terms of drivers, windows settings, exe properties, etc. However, it seems that maintaining 30fps, even with low settings, in a worst-case scenario like the A32NX in New York, isn’t possible with motion smoothing on. I maintain 35-45 with motion smoothing off but I can’t tolerate the juddering, especially looking at the yoke. I don’t suffer from any kind of motion sickness, it’s just a visual immersion killer for me.
When I go out into the countryside in a GA plane, it’s smooth aviation heaven…but like in 2D, I like to be able to set and forget. I like one configuration where I know I can hop into any plane anywhere in the world and get the performance I need. I set the sim for the worst-case scenario. I’ve ordered a Quest 2 to take advantage of its 18Hz ASW even if I do take a clarity hit. Even Valve’s Index (and the Vive line) now offer options to inject as much as six synthetic frames for every one real one. It seems that although the G2 is the killer headset to have and MSFS is the killer app to use it with, its ability to handle any scenario you can throw at it in the sim (within reason) is being let down by WMR’s inability to offer reprojection below 30fps. I wonder if there’s anything coming down the line that would expand on the reprojection abilities of WMR headsets and bring them more in line with what Valve and Oculus can already offer? @CptLucky8 Thanks so much for all of your work testing things and creating these kinds of threads, I had so much already done for me before I even took my G2 out of the box thanks to you.
I’m running a 5800X with 3080 and Reverb G2. Have just upgraded the CPU from a 3600X and thought I’d share the differences in performance, or at least my unverified observations.
TAA100/SS70, with mostly high settings up from 30-35 to 35-40fps, with LOD upped from 100-150. GPU bound, with plenty of CPU headroom. However, this is only with reprojection disabled. Reprojecion enabled gets me down to a MainThread bound 25, and exactly the same performance as with the 3600X.
I guess it begs the question - what’s causing the WMR OpenXR reporjection bottleneck? Sample size of 1, I know, but my experience says it’s not CPU limitations.
@LucioTheBird Thank you for your kind words and for sharing your experience!
@LucioTheBird@MickT1701
The reason you get poor results with motion smoothing is because it takes resources to compute!
I gauge this to about 25% of my frame time budget on my test system and this directly depends on your OXR/SteamVR SS: the larger, the slower. I also believe it is using maybe 50% CPU and 50% GPU but I’m not certain, it is just from the look of its effects on frame graphs and perf in sim.
Nevertheless, this is why I use the smallest OXR SS working with Motion Smoothing with the G2 (65% because of the OXR bug) and adjust the rest accordingly. In general, if I can render about 2K x 2K without, I can only render about 1.5K x 1.5K with motion smoothing with the G2, which is not bad still depending on the aircraft.
Hand in there though, there should be a new OpenXR preview soon with many fixes:
Thanks @CptLucky8 . I hope the brick wall isn’t hurting your head too much.
I get that reprojection doesn’t come for free, but I definitely can’t get that 1.5k vs 2k thing happening. The only way I managed to get a consistent 30fps was to fly in the countryside with every setting turned down to minimum, when it even managed to get to 45 fps. I then looked to work it back up using your bang for buck guide, but performance just fell off a cliff as I started to move from off to low.
Can’t see any cores getting maxed out when in reprojection, which suggests it’s waiting for something, rather than doing (much) more work. Have tried playing with g-sync on/off and CPL frame rate settings to see if they’re maybe having an unintentional effect, but no luck. Being able to hit 30/45 on stupidly low settings also counts against that theory.
My observations, and they are in no way as detailed or informed as yours, is that a significant increase in CPU power had no effect, and I don’t see any change in GPU frame times or usage when switching reprojection on or off.
Don’t know if the WMR OpenXR driver that actually handles the reprojection is open source - might have a look to see if it is, and dig around the code…
I have not much other choices with FS2020 and a 2070S, you should be able to get more with your 3080 though. However because of the actual FS2020 VR bottlenecks I wouldn’t expect you get more fps, but more “pixel shader” resources available. This means with the same settings I’ve published, you could get the same fps I get but with more clouds, more AO, more reflections, etc… (everything solely and directly using pixel shader)
You don’t see CPU/GPU rising otherwise because of the bottlenecks, and because FS2020 is mostly single-threaded and there is not much left ms available in the frame time budget to fill the slot with more computations.
I think I may end up putting the G2 in a drawer for a while when I get the Quest 2, I’ll keep an eye on the WMR software as it matures. Hopefully both of the major competitors (Valve and Oculus) having far superior motion smoothing options will motivate Microsoft to uodrafe the WMR motion smoothing features as well or at least optimise them somewhat. It’s good to know that there are fixes incoming for the 50 scaling bug, at least. Also, hopefully the optimisations that Asobo brings to the sim as they work on the console version will allow for more performance headroom. The sim being released on a console can only mean good things for the PC base.
I think the Quest 2 should he a good all-rounder for the sim for now until the G2 comes into its own. The fact that it allows for motion smoothing down to 18Hz and the fact that the chip in the HMD handles some of the processing for it should mean the performance impact with it on should be much less than WMR smoothing. I should have plenty of room to push the settings a good bit and still achieve a smooth experience and with a much larger sweet spot to boot, only sacrificing a little of the clarity of the G2.
Not actually chasing fps as such - just want to get 30 with reprojection on. In fact, with reprojection off, I just cap it at 30 via cpl and turn the settings up till I’m getting around gpu 28-30ms frame time. It does look glorious too
I have the same setup (5950x and 3090). Only switched SMT off with no additional PBO optimization yet.
I found that OXR on 150 with TAA 70 in the Sim and more or less all sliders on ultra with 200 LOD still gives me around 30 fps with everything else being crystal clear.
It feels like the 100/100 setting, but with better performance.
I don’t have a frame of reference other than reading the posts here. I got back into Flight Simulator after maybe 20 years, and had to build a new rig just for it. I would have bought a 3080 but it’s nearly impossible to do unless paying a scalper. A 3090 was tough to get at a fair price, but it was actually possible.
My requirements were to have a good VR experience, and so far have not been disappointed. At the moment my G2 is out of commission and am enjoying the sim with a UHDW monitor. I see the CPU (I9 10900K) would be around 35% while the GPU was 98 - 100%, and this is with all the VR settings but without the VR headset. Not sure any of this is relevant to the discussion in this thread but thought I would throw it out there in case this info is useful.
I am also about to exchange my 2070 Super for the 3090.
According to rumors, the 3080ti will come after the summer and then only have 12 GB.
Only the 3090 is clearly overpriced.
I’m also on a 2070S. I’m going to hold out for now.
Having upgraded my 3900 to a 5900, I was absolutely shocked at the improvement in performance just from the added cycles. Looking at some benchmarks, it seems that my seat-of-the-pants testing is confirmed, as they see 20-23% improvements in frametime going from 3900 to 5900, and only about 14-18% going from 2070S to 3080.
I can’t justify the 3090 at this point (and not entirely sure it could fit in my case, anyway :-)) – so I’ll be hanging on a while longer with the 2070s.
That is very interesting! I have an overclocked to 4.3Ghz and liquid-cooled 3700X. Benchmarks seem to show that 3900 to 5900 single core difference is ~17%. You are seeing a better result than that just from the CPU. That is an expensive CPU though. I’m wondering if it’s worth it to upgrade 3700X to 5600X which has nearly identical single core performance (just less cores). I thought it’s too expensive for the performance difference, but if single-core or multi-core performance difference translate to a larger FPS gain, I may reconsider…
I should note that I think its NOT necessary to get a 5900 for a big improvement. In your case a 5600 or 5800 is great (I just snagged a 5900 when I saw it, and overspent for my needs.)
Do NOT go 5900 on my recommendation – any 5x00 is going to be a step up. That said, YMMV. I wasn’t doing an OC, just running SMT off, on my 3900.
3700X vs 5600X is +15% single core performance (which should be most indicative for MSFS as it’s not great with multithreading). Similar to your to 3900x vs 5900x +17%. The difference from your 5900X to 5600X in single-core performance shows 3%, confirming the above. I was wondering if it would translate to a significant real-world difference in MSFS, especially considering all I’m seeing is “Limited by GPU”. In your case you’re saying 17% translated to 20-23% difference, which is significant. I also have 64Gb of 2600 RAM overclocked to 3200 but with slower timing (21). It’s original timing is 16/2600 and 21/3200 gives the best speed of anything I could make it run with. I was thinking of upgrading to faster RAM - 3200 native or faster, with faster timings (14?). That’s a lot of money… I wonder if the performance improvement will be worth it…
Your guide improved my overall 2070S to 35/36 fps. One thing I did get, after following the guide, was an issue with the WMR Virtual Monitor registry edit.
I would get a CTD every time I exited VR to go back to FS2020 in pancake mode. I went back step by step with drivers, Nvidia settings etc and the culprit was the ‘prealloctevirtualmonitors’ registry entry. Once deleted, the CTD at VR exit dissapeared.
Just in case this helps anyone who sees the same problem.
Just kidding.