VR settings For AMD Navi2 Based systems (6900xt/3900x/G2)

[nikoel]

Hey guys. So first of all, I’ve created this thread based upon my experiences and knowledge in DCS and around AMD Graphic Card/Processors. Setup is quite a sensitive thing, so straight out of the gate let me pre-phrase with the following: What I write will be set for my 3900x, 6900xt and my Reverb G2, with everything overclocked to the tits. Whilst I will post my settings and screenshots, with updates they inevitably will change. A lot of it my opinion which might contradict your experiences or opinions, and that’s aye okay!I am complimenting @speed-of-heat ‘s amazing settings and comparison guide here: https://forums.eagle.ru/topic/257819-my-3090-settings-for-my-g2-279/

And of course Thud’s https://vr4dcs.com blog & discord. If you haven’t already, go there and read them, and come back to this dumpster fire later

 

What am I trying to achieve: The best quality settings with smooth gameplay and no stuttering/ghosting when looking at a bandit at 9/3 o’clock. Nothing more, nothing less.

 

In addition I am also making this thread to share a little bit of knowledge which is unlikely to change to disprove some myths that seem to floating around which sent me down a pathway I don’t wish on anyone. I am also hoping that this thread will be deprecated with the much anticipated new arrival of Vulkan and new implementations in EDGE 2.7. 

So, the whole VR thing is a mess. Settings are everywhere. 

Like a 3 year old kid in a supermarket, grabbing a Toblerone at the bottom of a delicately stacked pyramid in a shape of a Christmas tree - ticking one wrong box can bring the entire otherwise perfectly stacked creation that took hours of testing and re-testing to a grinding crash [and lead you down the wrong rabbit hole only leaving you wishing you could go back when things were mediocre but you’ve changed so many things now, you wish you could remember what those settings were to begin with

 

Maybe you will start your journey like I did. With ghosting and stutters but an okay-ish image - so you turn to advice and turn on ASW/Reprojection. This tanks performance. Thus you to turn lower the resolution to get some of it back. Then of course, upscale. But this doesn’t look good. So we are adding additional overhead to sharpen the image, and then a bunch of shaders to try to do something about the mess that is appearing on the screen Then add MSAA and AA to complete the moving low poly water colour painting within the powerpoint slideshow. Perfection! We now have a worse image, with smear across the screen, but hey! At least it’s ASW consistent and the GPU is running at 60% with “overhead”… yay!

 

To see where I was going wrong, to keep things simple and this post reasonably short we must understand three things. 

1) Resolution 2) Antialiasing 3) Upscaling 4) Which settings are “worth it” and which goals are worth not striving for as of writing this (Feb 2022)

 

1. Let’s quickly talk resolution inside SteamVR and why the values are so confusing. The resolution at 100% is significantly higher than the native resolution of your headset. In short, the system must render a picture larger than the physical size of the screen to counter the ‘pincushion effect’ created by the lenses. The recommended resolution from Valve is 1.4x the display resolution of the screens. This is why even though say the Valve Index has 1440 x 1600 resolution screens, 100% render resolution in SteamVR is 2016 x 2240 per eye

 

PD Values, Render resolution and per app steam resolution etc etc etc… all in principle achieve the same thing. You can chose to keep it simple by leaving everything at 1.0 and 100% and then manipulating one slider to change the end resolution (whichever you want – but I recommend ‘per application’ one inside SteamVR) - or you can do your voodoo bomfire dance around your computer by turning SteamVR to 500%, Per application to 50% and PD to square-root of 0.90 and pretend you’ve achieved something meaningful when in reality leaving the sliders at their default 1 and 100 would have given you the same results. You can not cheat your way out of rendering pixels. You’re just moving multipliers around to end up at the same rendering value and making things super confusing and complicated as the multiplicators within different apps all work in different ways. (The only exception to this rule is the FSR/NIS mod which I will get to)

 

2. Antialiasing. In short, It turns jiggered edges into smooth edges by trying to guess and smoothen (and often blurring) lines. (As all simplifications, this is slightly wrong but good enough explanation for us here). This is especially important at low resolution where say a wire from a mast would be only a few pixels wide. Thus you get this weird diagonal zigzag. This becomes less important at higher resolutions as that wire has significantly higher pixel count and will appear smoother. (I still remember an episode from LTT when he rendered a car racing game at 8k. He had antialiasing completely off. Because at that resolution there were no jiggered edges due to the ridiculous amount of pixels and polygons being rendered) There are many ways to do this, they mostly involve acronyms eg TAA, MSAA etc… You can google them and see what they do for yourself. In my experience, DCS’s implementation of antialiasing is kind of “selective”. In that it produces good results inside the cockpit and horrendous results for things like buildings and far away objects such as clouds. Weird.

 

Why is this important? Because in DCS the 6900xt can render like the A10c can brrt. And just like an A10c in a dogfight, It falls over flat on it’s face as soon as soon as we go box ticking. Those boxes, in no order are, ASW, AA, MSAA, FXAA, Reprojection; heck, pick your letter combination and it will drain performance. In my testing against a 3080ti, those things affected AMD cards a lot more than Nvidea I have no idea why. But, thankfully that’s okay

 

3) Currently the best upscaler is by Nvidea and it’s called DLSS 2.#. It’s not available for DCS or AMD cards altogether. Lets move on. Steam VR uses an advanced upscaler. In my testing it produced the worst results. This basically leaves two others, AMD FSR and Nvidea’s NIS. These use advanced algorithms upscale the image and do much more than a simply blow up of the image and add a sharpening filter on top (I will get to those as well, spoiler alert also ugh! But I get it). We have those available to us via a mod – two files to be exact! https://github.com/fholger/vrperfkit Thank you fholger! 

 

For FSR to work well it needs as many pixels as you can render! I have tested many configurations and I found the sweet-spot between performance to be around 0.7-0.8 with the render resolution of 115-130% in steam VR for my Reverb G2. It will struggle to upscale if you are not giving it enough pixels to upscale. It will look like crap and it give you artifacts and meh results. It’s not a miracle creator

 

A word on sharpening filters, shaders and other overhead snake-oil. Use them if you like the result. In my testing a lot of the shimmers are caused by these sharpening filters. I could only conclude that provided you are rendering high enough resolution all these additional "features" are meaningless and cause problems. Play to the card's strengths, and stay away from weakness’ – they have a small overhead though. I have tested the latest Resahde V5 (as of time of writing this) and the filters, shaders are amazing – the filmic sharpening filter is awesome. But in the end the best performance and image quality that eluded me for so long finally was there when I began to understand these simple things

 

Now you’ve read this far. You’ve earned your cookie. In case it’s not already obvious. We increase settings where graphics card is strong and stay away from settings which bring it to it’s knees. With other words we need to crank the resolution and let the poly-count do the antialiasing work. Then crank the resolution even higher and use AMD FSR to upscale to try to keep shimmering and jiggered edges to a minimum. You will have shimmering (especially long distance clouds) and you will have some sharp edges inside the cockpit. But you won’t need a sharpening filter which also reduces performance. This is a personal preference, however imho it’s a trade-off that is worth it 100 times over. Avoid ASW, avoid MSAA avoid AA and all additional gyzmos and overheads, put all of your cookies into the resolution basket and let it rip

 

With 6900xt Its easy to make the game look smooth and good when you’re looking straight ahead. The real test is looking out the side (9/3’o clock)

 

I personally render my G2 at 115-130% and with frametimes often below 10ms. This is over 3.5k resolution per eye, I can read the switch labels in the F-14 (yes those labels) and there is no smearing or ghosting when looking at a bandit on 9/3 o’clock. You also notice how I don’t say what FPS – it doesn’t matter. Once over at or over 45ishFPS (will vary from person to person), forget about the fps and concentrate on frame times and consistency and how it feels. To render at 90hz (G2 rate) you need timing of 11.1ms, thus this is a guide of where the perfect frame times are. Chances are through, you won’t be at 90fps with the settings above

 

A quick word on your processor speed. I have overclocked mine per CCX. This means that all cores are locked at what I consider their maximum speed. (For me the fastest sits at 4.5Ghz and slowest at 4.35) I generally had bad experience with PBO due to it’s tendency to boost to 4.6Ghz and then drop down and boost again thus giving me irregular performance. This affects DCS a lot more than other games in that it uses a small number of CPU threads. 

 

Settings and things I had to accept. In no particular order. Even though I am super sampling like crazy, Jiggered edges are and will be a thing. Distant clouds in some lighting configurations are a disaster. They are not worth fixing. The amount of work needed to get them to shimmer even a little less will bring performance to it’s knees. Buildings shimmer. I live with that too. Dropping cluster bombs will bring your FPS to side-show like levels. It’s a bug that hopefully will be fixed, don’t drop them for now. FSR will make distance objects look a little weird at times. Accept lower framerates when on the supercarrier. I am not lowering my FPS to a fixed setting, thus fluctuations are a thing

VR has been and is a battle of 1-3% improvements per step. Each combine to give you a sizable performance improvement in the end. There is no magic bullet. A good place to start is the excellent blog by Thud - https://vr4dcs.com

Here are my settings – just plugging these in blindly is unlikely to give you the optimum performance for your setup but might be able to get you started

Edited February 11, 2022 by nikoel

[speed-of-heat]

Quote

For FSR to work well it needs as many pixels as you can render! I have tested many configurations and I found the sweet-spot between performance to be around 0.7-0.8 with the render resolution of 115-130% in steam VR for my Reverb G2. It will struggle to upscale if you are not giving it enough pixels to upscale. It will look like crap and it give you artifacts and meh results. It’s not a miracle creator

I will be giving that a go later thanks for the tip!

[DeltaMike]

Excellent writeup.

TED shows us that the 6900XT's performance is really close to 3080 -- within margin of error it seems -- and given the prices out there, I imagine we will be seeing a lot of 6900XT/G2 systems coming on line in the coming year.   Threads like this are invaluable to new users.  Couple of comments:

1.  Overclocking AMD GPU's isn't as easy as it sounds -- those things have a mind of their own.  But it's a pretty good mind.  Best advice to a noob is, keep it cool.  Every cooling strategy starts with fans.  As Lynyrd Skynyrd says, "Turn it up."  There's nothing good about a quiet rock n roll song, or AMD GPU.

2.  Other than fan speed, the most important setting in Adrenaline is your minimum GPU clock, which must be set a couple hundred less than your maximum.  Keep your GPU clock in a narrow range to avoid frametime spikes.  

3. There is no substitute for pixels-per-degree.  Back in the Oculus Rift days, we twisted ourselves in knots trying to turn a blurry mess into a playable game.  Supersampling, MSAA. It's a wonder my old Vega 56 didn't catch on fire.  We don't have to do that with the G2, and maybe we shouldn't.  Agree, people should at least try running at native resolution without anti-aliasing and see if they like it.  

4.  AMD handles up-sampling pretty well.  In other words, if you must undersample, the game still looks pretty darn good.  I'm not sure I can perceive a difference between running the FSR mod at 0.8 vs setting SVR to 80%, just flying around.  And SVR at least can be adjusted "on the fly."  (Theoretically you could undersample while you're on the deck, and then crank it back up again when you fence in so you can spot those pesky MIGs)  

5.  It's not totally clear to me one accomplishes anything by running SVR at 120 and FSR mod at 0.8.  Assuming those things add up, 1.2*0.8=.96.  See #3 above. I'm not saying the FSR mod is superfluous -- haven't quite made up my mind yet.  Let's just say it's *really* close.  

6.  Anisotropic filtering is a freebie and it really helps a lot.  Crank it up.

7.  In my opinion, MSAA has aesthetic benefits but it doesn't enhance game play.  Especially for ACM.  Your best spotting distance will be native resolution with MSAA off.  Second best is FSR mod with MSAA on, which works OK but it's harder to spot a blob than a dot, except against a clear sky.  Undersampling with SVR alone will hurt your spotting distance (because of "pin cushion" or barrel distortion).   

8.  "Stutter" can either be regular or irregular in rhythm.  (eg disco is "regular", breakbeat is "irregular").  Irregular stutter can be reduced by tuning your card and adjusting DCS settings.  I believe locking your frame rate at 30, 60, 45 or 90 will also reduce irregular stuttering.  Regular stuttering -- what film makers call "judder" -- is pretty much inevitable.  Fly at 500 feet, 500 knots, roll over and look straight down at the ground and you'll see what I mean.  Motion reprojection will reduce that, at the expense of introducing artifact in text (eg instruments, DDI displays)

To a certain extent we need to pick our poison.

-- For ACM, I'd suggest getting rid of the eye candy, run at native resolution, and turn MSAA off and motion reprojection on. 

-- For flying at night, with your head down in the cockpit, I'd turn motion reprojection off and maybe lock frame rate at 45.  

-- For watching the sun go down over the desert, I'd max out shadows, undersample to about 80%, and turn MSAA on.  So pretty! (Although I rather imagine MSFS might be prettier, we'll see)

That's for jet jockeys of course.  Warbirds and helo's have different needs I imagine.  

 

[nikoel]

2 hours ago, DeltaMike said:

Excellent writeup.

TED shows us that the 6900XT's performance is really close to 3080 -- within margin of error it seems -- and given the prices out there, I imagine we will be seeing a lot of 6900XT/G2 systems coming on line in the coming year.   Threads like this are invaluable to new users.  Couple of comments:

1.  Overclocking AMD GPU's isn't as easy as it sounds -- those things have a mind of their own.  But it's a pretty good mind.  Best advice to a noob is, keep it cool.  Every cooling strategy starts with fans.  As Lynyrd Skynyrd says, "Turn it up."  There's nothing good about a quiet rock n roll song, or AMD GPU.

2.  Other than fan speed, the most important setting in Adrenaline is your minimum GPU clock, which must be set a couple hundred less than your maximum.  Keep your GPU clock in a narrow range to avoid frametime spikes.  

3. There is no substitute for pixels-per-degree.  Back in the Oculus Rift days, we twisted ourselves in knots trying to turn a blurry mess into a playable game.  Supersampling, MSAA. It's a wonder my old Vega 56 didn't catch on fire.  We don't have to do that with the G2, and maybe we shouldn't.  Agree, people should at least try running at native resolution without anti-aliasing and see if they like it.  

4.  AMD handles up-sampling pretty well.  In other words, if you must undersample, the game still looks pretty darn good.  I'm not sure I can perceive a difference between running the FSR mod at 0.8 vs setting SVR to 80%, just flying around.  And SVR at least can be adjusted "on the fly."  (Theoretically you could undersample while you're on the deck, and then crank it back up again when you fence in so you can spot those pesky MIGs)  

5.  It's not totally clear to me one accomplishes anything by running SVR at 120 and FSR mod at 0.8.  Assuming those things add up, 1.2*0.8=.96.  See #3 above. I'm not saying the FSR mod is superfluous -- haven't quite made up my mind yet.  Let's just say it's *really* close.  

6.  Anisotropic filtering is a freebie and it really helps a lot.  Crank it up.

7.  In my opinion, MSAA has aesthetic benefits but it doesn't enhance game play.  Especially for ACM.  Your best spotting distance will be native resolution with MSAA off.  Second best is FSR mod with MSAA on, which works OK but it's harder to spot a blob than a dot, except against a clear sky.  Undersampling with SVR alone will hurt your spotting distance (because of "pin cushion" or barrel distortion).   

8.  "Stutter" can either be regular or irregular in rhythm.  (eg disco is "regular", breakbeat is "irregular").  Irregular stutter can be reduced by tuning your card and adjusting DCS settings.  I believe locking your frame rate at 30, 60, 45 or 90 will also reduce irregular stuttering.  Regular stuttering -- what film makers call "judder" -- is pretty much inevitable.  Fly at 500 feet, 500 knots, roll over and look straight down at the ground and you'll see what I mean.  Motion reprojection will reduce that, at the expense of introducing artifact in text (eg instruments, DDI displays)

To a certain extent we need to pick our poison.

-- For ACM, I'd suggest getting rid of the eye candy, run at native resolution, and turn MSAA off and motion reprojection on. 

-- For flying at night, with your head down in the cockpit, I'd turn motion reprojection off and maybe lock frame rate at 45.  

-- For watching the sun go down over the desert, I'd max out shadows, undersample to about 80%, and turn MSAA on.  So pretty! (Although I rather imagine MSFS might be prettier, we'll see)

That's for jet jockeys of course.  Warbirds and helo's have different needs I imagine.  

 

Naw - thank you, and for your additions as well

I'll try to answer a few questions

FSR is a significantly better technology, for it to work it requires resolution to upsample. The fewer pixels you give it, the worse the result. In other games, it's value lies in taking an image that is around 1440 and turning it into 4K. Even at 1440 there is some merit in the technology. It struggles with image quality below that. So depending on what headset you have and what resolution you're using, I agree there may not be much of a difference between it and normal SteamVR upsampling. This is why I believe so many people on the forums disagree on whether it's a good thing for them,

5 & 6. So, It accomplishes a better image. Resolution is not the whole story (although rendering above native still helps, because of FSR/ASF) - It's better becuase we are rendering more polygons. This is a form of anti-aliasing; we are basically supersampling but instead of a tick box, we do it with fine control. Best of all, its the absolute best visual kind of AA; which will outperform in quality every other kind in existence. Thus the image is smoother and nicer to look at. Then we utilise AMD FSR and Steam's Advanced Supersample Filtering to reduce shimmers. This works because the two technologies work in unison. First FSR filters out the random pixels from the image because it uses the pixels and the entire image to "guess" what should be there; and then SteamVR's ASF is able to apply an other layer of rejection where it looks at extra pixel and averages out the result (simplifying a lot here). The reason you see AA turned off is because it still adds a bit of overhead, and for me, because of the above, it produces a worse image. So what I am saying is that once you can render above 100% I believe you can have your cake and eat it too - but alas there are definite concessions

8 At Mach speeds in viggen, flying low level thankfully I don't get any of those effects and able to look at the trees without them appearing irregular or somehow choppy. What I was talking about is the tendency of the aircraft at close proximity at 3/9 o clock position to smear, jump and ghost. This was especially annoying during the merge and one circle fights when using the vertical and seeing them smear across the canopy. Now everything is fluid

 

[TED]

Good write up @nikoel. It’s great to see more people testing and tweaking and it’s clear to me there are different routes to go to get great performance. I agree with a lot of what you say and may well give a go. For now I’m a big fan of motion vector and motion smoothing so tune to that. I totally agree with the pixel count being all important. My view is very much to get it up to or as close to native resolution as possible then implement fsr to allow some other benefits and msaa. 
 

I have both a 3080 and 6900xt and in my view the 6900xt is a clear winner in dcs although it took a bit more tuning and rethinking of the logic. But in the end I was able to get considerably better, smoother and more pleasant performance from my 6900xt, mostly with a10c. There are of course huge differences from terrain to terrain and multiplayer often requires some other sacrifices and there it is quite personal on what you fly and what you are and are not willing to let go. 
im taking a break from multiplayer for now and focusing more on campaigns until we see multicore and Vulkan allow us to squeeze some more out of the CPU’s in high unit terrains. 
 

with native resolution on my g2, mv and motion smoothing, msaa x2 it’s very smooth even looking out the side at low level. Most dcs settings high, but grass and clutter off, shadows flat. With openfsr applied it’s really excellent performance. 

[DeltaMike]

It appears FSR scales like pixel density, so we can plug it in to jabbers' spreadsheet to see if we are making progress or not, performance-wise. 

I think performance is gonna correlate with total pixels rendered, but it is very interesting how you get there.  Essentially using supersampling as your anti-aliasing technique, no?  In other words, take care of those jaggies *before* fsr upscales the image.... 

Very interesting approach and I'm eager to give it a try.  

[TED]

So I tried your approach to this with mv and motion smoothing off and force dx11 off. 

For me it's just doesn't work for some reason. Too many small stutters. The force dx11 in wmr seems critical. I've no idea why. With it checked off the image is better but more stuttering. 

So I've had to tune for that too. 

Motion vector and smoothing is just too beneficial and it doesn't work properly if dx11 isn't forced on. 

However I bumped svr res to 132% textures high, but msaa off and it's actually an improvement. I needed to drop resolution to 100% or less to use msaa reliably. However msaa softens the image. Without it but at highest resolution it performs better and the image is generally better (subjectively). 

Lots of good stuff coming out of all these threads. 

Id love to know what the dx11 effect is though and why it seems to have such a big effect on dcs with amd cards. If I could get mv to work without forcing dx11 it would be perfect. Perhaps someone cleverer than me can speculate. 

[DeltaMike]

Running this through Jabbers' spreadsheet, where "effective PD" is a measure of how many pixels total we are driving.  (Actual in-game PD is kept at 1.0) F18, cauc free flight instant mission. 

Open FSR stock:  SVR 100%, multiplier 100%, FSR downscale value 0.7, MSAA on.   100/100/1 refers to bone stock native resolution without the FSR mod, MSAA off.  Final entry is: SVR 100%, multiplier 120%, FSR downscale value 0.7, MSAA off. 

MSAA just crushes performance.  Clearly way better off using nikoel's strategy.  I wouldn't say it's perfect in terms of getting rid of the shimmering, although it's about the same as running native resolution and I feel it looks better than 100/100/.7 even with MSAA on (although I think I'm getting a little too aggressive with the sharpening).

Note, spotting distance (for an S3) is about 8 miles in native resolution, and is less than five miles at 100/120/.7.  MSAA off for both trials.  

Still not completely sold on FSR although if ED sees fit to implement it -- with, for example,  the HUD and DDI's coming  through at native resolution -- I could see the utility

 

Edited February 11, 2022 by DeltaMike

[TZeer]

Did some testing again now.

Wow, the FOV trick was a massive performance gainer. I did not notice any loss of FOV physically inside the oculus set.
I put them at 0.8 on vertical and horizontal.

My setup:
6900XT OC @ 2650Hz
GPU memory default

5800X @ stock settings
32 Gb Ram

All DCS settings pretty much as in  @speed-of-heat  excellent thread , but MSAA OFF, visibility to extreme and scenery details factor 0.8.
FSR installed , default settings.
ASW OFF

Rock solid at 72 FPS except for a small second where it dropped to 66FPS in the turn around the hill inverted over the forrest. But hardly noticable.

So during the whole test track, it was buttery smooth, no artifacts and the GPU still had some headroom to play. Kept the load between 82-95%.

Apparently what taxes the card the most is close up to forrest while flying inverted. So all I see is forrest.

I love how much knowledge there is out here when it comes to tuning the cards and settings in DCS. It's just a matter of compiling it all and getting it out to the masses.

Edit:
Flying in Nevada during sunset must be a dream now. Hardly any forrest, and some headroom to increase the resolution a bit more. 
Now I just need to find some sort of shader that can fix the colours and works with oculus. Looks kinda washed out. 

 

Edited February 12, 2022 by TZeer

[DishDoggie]

Without a doubt one of the best VR posts. Thank You for this post @nikoel 

 

This Video Post was not made by me but I use it all the time after a Update. I would like to add it to the post just because of the Mask removal of the output screen so you can record the VR Flatscreen in full without a round VR mask for posts like this and other needs.(In VR Videos). I Book Marked This in my VR tab you should too. (along with this Post by "nikoel" Thanks to @plazma for making this.

 

[edmuss]

Been playing with SVR 120% FSR 0.75 with no sharpening at all and it's giving a decent image without the hit of MSAA.  Good enough for mostly over 60fps over syria on an empty mission with medium high settings.  A10C easy instant action is 12-16ms frametimes for the duration of the mission.  Running at 60Hz so smoothing is generally disengaged which is ideal.
Only other mods are Kegety/SoH's clearwater and Taz's low bit depth A10 normal maps.

Logical now that you've explained it @nikoel so thanks for that

[nikoel]

Hey guys [and girls?] 

Thank you so much for the kind words. I was a little hesitant writing the thread and was preparing myself to get flamed for proposing such ridiculous settings - Im glad it's working out for some of you. I do want to stress why this works for me so well and may require further tweaking on your end. It's all personal, and these settings are not perfect. Much like most things in life they are a tradeoff. Just in my opinion that tradeoff is miles ahead of anything else I have tried so far

1) I am using AMD FSR for a combined render resolution of 7K (3.5k per eye on my G2) this means that the technology has an already great image to upscale further. This is a really frustrating fact of life and how FSR works, you already must be able to render "good resolution" before you start. I don't know how much this well help if you have a card with performance less of say 6800/3070/2080. Judging by @edmusspost above, it might work well for Nvidea as well? 

2) I am overclocked very high, my 6900xt is sitting at 2.8Ghz and 360W power envelope with fast timing and VRAM at 2150mhz, the processor whilst not exactly new, is also locked at 4.5ghz on one of the CCX. It takes multiple attempts to post into windows due to power/load-line calibrations I have implemented. I have also gone through all the windows tweaks published and even altered some registry's to potentially maximise performance there as well. Hell knows if it actually works, still perplexes me how Microsoft would try and "hide" performance away from users via a couple of toggles and registry edits. But what do I know. I also give out of my way to quit all apps (that in theory shouldn't make a grain of difference since 90% is idling when playing DCS anyway). You may need to lower settings/resolution depending on where your system sits in comparison to the one above. On the G2 with the above settings you should be able to see other traffic from around 40nm away provided the jelly clouds are not screwing things up for you 

3) I messed with DX11 toggle. I concluded that I was getting better performance via the DX11 toggle but the image quality is significantly worse. This is where the performance gain comes from. If the quality is already low (whether via your settings or your headset) it's more difficult to see the difference. I don't know why this happens. Someone much smarter than me may be able to answer this

4) This technique needs SteamVR Advanced Supersampling. This a copy and paste from what it does: Anything more than 4 neighboring pixels being downscaled to 1 will cause aliasing with a single bilinear interpolated sample (because underneath it is actually mixing four samples). This means the periphery will be undersampled when things are at ideal res. So advanced supersampling adds in more samples in these areas so that there is less aliasing and so that the rendered data there isn't thrown out. This is super important if you're rendering a resolution as high as possible

Edited February 13, 2022 by nikoel

[edmuss]

I have DX11 toggled on and haven't tested the 120/0.75 without, I will do and see if it makes any difference. I already had the advanced supersample turned on.

Edited February 13, 2022 by edmuss

[speed-of-heat]

So there are definite FPS to be had with this approach, but, my observation is I get a less "shimmery" look with MSAA x2 SS 100 and CAS at 0.80 and less jagged than than by over sampling to 150 and down sampling to 0.8 with CAS (which should be a similar number of pixels);  now shimmer is in the eye of the beholder, but it honestly isn't for me  that doesnt make it wrong... just not right for me.

[nikoel]

50 minutes ago, speed-of-heat said:

So there are definite FPS to be had with this approach, but, my observation is I get a less "shimmery" look with MSAA x2 SS 100 and CAS at 0.80 and less jagged than than by over sampling to 150 and down sampling to 0.8 with CAS (which should be a similar number of pixels);  now shimmer is in the eye of the beholder, but it honestly isn't for me  that doesnt make it wrong... just not right for me.

Thanks mate, did you mean FSR? CAS is Contrast Adaptive Sharpening. I was using FSR without any sharpening because it has advanced upscaler that CAS lacks and in my testing sharpening added shimmer. I believe CAS would have added very pronounced edges giving you that Borderlands 3 look, especially cranked to 80%

Edited February 13, 2022 by nikoel

[speed-of-heat]

1 hour ago, nikoel said:

Thanks mate, did you mean FSR? CAS is Contrast Adaptive Sharpening. I was using FSR without any sharpening because it has advanced upscaler that CAS lacks and in my testing sharpening added shimmer. I believe CAS would have added very pronounced edges giving you that Borderlands 3 look, especially cranked to 80%

 

I meant CAS, i will give it another shot with FSR, btw no borderlands effect 

Edited February 13, 2022 by speed-of-heat

[speed-of-heat]

OK, FSR makes a HUGE difference from a visuals point of view. more testing required now

[edmuss]

I found that CAS at the same upscaling as FSR gave a good bump in FPS but wasn't as clear, things looked a bit indistinct and with more shimmer; MSAA was required to calm it down which negated the performance boost.

For me FSR gives a clearer image but upscaling from the higher initial resolution reduces the need for MSAA with minimal performance hit.

Curiously 50% resolution only gains me about 2ms on 100% but at significant loss of clarity, MSAA on top of that costs more than the saving and it still looks a bit rubbish. Running 120% over 100% is only a relativity small performance hit.

edit1: tried turning off the DX11 toggle and it breaks my motion smoothing entirely, both WMR and steamvr. It does appear to give me a couple of extra ms though but even though I'm below 14ms with a green framegraph it's jerky as hell. I tried with HAGS on (my normal setting because it alleviates the vram hole bug) and off and both result in the same thing.

Shame really as the extra performance would be handy, will do more testing to see in there's something in nvcpl I have that might affect it. For what it's worth I saw no difference in visual quality between DX11 forced on or off.

edit2: I'm not sure what settings have made the effect but I now have force DX11 toggled off, HAGS off and smoothing handled by steam rather than WMR.

Pushed it up to 130% with FSR 0.75 and now getting 14-15ms where was getting 18-19ms with 100% FSR0.75 and MSAAx2.  Granted the road edges are still visibly dancing but on the whole it's a much more stable image.  500 foot over downtown Beirut is 15-21ms and 4000 foot over the Bekaa valley is 14-16ms.

Unfortunately now I have HAGS off I'm suffering more with the vram hole, hopefully that bug gets squashed soon.

Edited February 13, 2022 by edmuss

[nikoel]

10 hours ago, speed-of-heat said:

OK, FSR makes a HUGE difference from a visuals point of view. more testing required now

So a little bit of Borderlands then :P. I kid kid

Can I ask a favour? Are you able to explain which shader mods you're running? I had a look at the ones in your links but holy Jesus, the threads are a mess. There are mods on top of mods, which rely on tweaks that others have made to make them IC compliant or fix them after an update. I have no idea which I should install and in what order. I run OVGME but the more I look into it the more confused I get

If you wanted to fly multiplayer with IC would you install any of them. If so which one specifically?

 

[speed-of-heat]

Simplex… not perfect but work well enough. https://www.digitalcombatsimulator.com/en/files/3317762/
 

[DeltaMike]

Some more tests.

1.  Steam thinks I should run at 132%, which is ~3600x3600 virtually rendered.  Beautiful image, virtually no shimmering, as supersampling is used as anti-aliasing.

2.  If I dial pixel density down to 0.8, end of the day I'm rendering at about native resolution.  There is some shimmering but my times are good and the game is playable spotting-wise  This is where I usually fly (my actual settings are 100/100/1.0.  More than one way to skin this cat).

3.  I installed vrperfkit and ran it at suggested settings all the way around (fixed foveated rendering is off, of course).  Likewise it winds up being about native resolution, so there isn't really any net undersampling.  The shimmering is there, so there's no real anti-aliasing either.  All I'm really using here is the sharpening filter, which I will admit, looks kinda nice and didn't cost me anything.  (I should probably select "CAS" rather than "FSR," I guess)

I think the thing works as advertised -- FSR at Ultra is virtually indistinguishable from native resolution, and it runs a heck of a lot faster.  Doesn't add much to the 6900xt unless you like the sharpening filters, which are starting to grow on me. 

[nikoel]

20 hours ago, speed-of-heat said:

Simplex… not perfect but work well enough. https://www.digitalcombatsimulator.com/en/files/3317762/
 

Thank you. Gave me an other what feels around 10ish% boost. F14 now feels like F18 did before the mod. 

@DeltaMike- What graphics card do you have? Is it still the Vega 56 in your signature; because if yes, this is amazing performance for something that is a couple generations old. I gather by the resolution you are using a Rift S?

[DeltaMike]

oops gotta change that, whole new system with 6900xt

that vega punched above its class but that was then...

OK, let's compare apples to apples:

1.  SVR 100%, multiplier 100%, PD 0.9, vrperfkit off

2.  SVR 132%, multiplier 100%, PD 1.0, RenderScale 0.77

Both function and look exactly the same, which is what we would expect -- we are driving the same number of pixels.  GPU time 11.8 either way.  Moderate amount of pixellation and shimmering, DDI's are reasonably clear.  

Key to the whole banana is to run #2 twice -- once with the advanced supersampler filter on, and once with it off.  I did that, and I think it might help a little in terms of pixellation and shimmering.  Maybe.  

It's clear to me that undersampling to any appreciable extent -- a PD or RenderScale of less than about 0.7 -- makes the picture look awful.  Terribly blurry and pixellated.  Everything looks more or less OK if you only undersample a little, and it's open to debate whether vrperfkit adds much of anything beyond the sharpening filter.  

Question is, to what extent does the advanced supersampling filter work.  If it's working as we hope and suspect, then cranking up the steam resolution settings may help, even if the total number of pixels rendered is the same.  

We need to go back and really crank it up.  Reproduce the Jabbers experiment, keeping total pixels (or "effective PD") constant and only vary the advanced supersample filter.  Looking not so much at GPU time -- which we expect to be the same -- but at jaggies and shimmering.  

Worth looking into I think.  I've noted that a moderate amount of oversampling (ie resolution at 132%) doesn't affect spotting distance.  Undersampling and then applying MSAA destroys it.  

 

Edited February 15, 2022 by DeltaMike

[TED]

I’ve been trying various settings and also replicated exactly @nikoelsettings. I suspect every system is different but for me without the motion vector and motion smoothing there is just too much micro stutter. 
It may even be a bit module dependent but I need to force dx11 and have mv and smoothing on, then everything is really smooth, even in a gazelle at 50ft through valleys and trees or over cities. 
Most of the rest of the settings however are the same or very similar to Nikoel. Svr 100%, per app 120%. I’m using fsr with upscale of 0,9, radius 0.8 and sharpness 0.85. 
msaa x2 works fine without any real hit in single player but it’s not really necessary with the resolution dialed up. 
I think a lot of this is a bit subjective but I just hate even tiny stutters and without the smoothing it’s still noticeable for me. 

Edited February 20, 2022 by TED

[nikoel]

6 hours ago, TED said:

I’ve been trying various settings and also replicated exactly @nikoelsettings. I suspect every system is different but for me without the motion vector and motion smoothing there is just too much micro stutter. 
It may even be a bit module dependent but I need to force dx11 and have mv and smoothing on, then everything is really smooth, even in a gazelle at 50ft through valleys and trees or over cities. 
Most of the rest of the settings however are the same or very similar to Nikoel. Svr 100%, per app 120%. I’m using fsr with upscale of 0,9, radius 0.8 and sharpness 0.85. 
msaa x2 works fine without any real hit in single player but it’s not really necessary with the resolution dialed up. 
I think a lot of this is a bit subjective but I just hate even tiny stutters and without the smoothing it’s still noticeable for me. 

 

Yeah, fair enough. DCS is weird like that, people with exactly the same hardware seem to get extremely different results whilst running exactly the same settings. It's just an other method that obviously works for some and does not for others. I think @speed-of-heatstarted recommending 140% in his latest post on his nvidea recommendations page

I would force DX11 on as well. The performance gains are awesome; but I can not stand the loss of clarity after being able to read everything in the F14 and so it's staying off