Jump to content

G2, supersampling, MSAA performance face-off


DeltaMike

Recommended Posts

Observations:

1.  MSAA gives the best result in terms of reducing shimmering in the cockpit, on shorelines, and on runway edges, and also does the best job of defining structures such as cranes, bridges and powerlines, at the approximate expense of running 150% resolution

2.  Supersampling using steam settings has an intermediate effect on performance and cosmetics.  Significantly reduced shimmering in cockpit textures, somewhat reduced shimmering along shorelines and runway edges, and a modest improvement is the resolution of ground structures such as bridges and power lines.  Not as good as MSAA, but not as expensive performance-wise, and considerably more granular.  

3.  It does not appear that the DCS setting "SSAA" has any effect on performance, or visuals.  Not clear that it is doing anything in VR.

4.  Likewise, whether or not Steam Advanced Supersampling is applied does not appear to affect performance, and does not appear to affect shimmering in cockpit textures, shorelines or runways.  However I got the feeling it may result in better definition of ground structures such as bridges and powerlines.  It doesn't cost anything, so I suppose it's worth trying.  

 

DISCUSSION

Testing was done with a 6900XT which can presently run the G2 at native resolution (100/100/1.0) with little difficulty, however it results in a very "hard" image with considerable shimmering evident along edges.  I'm assuming this is due to pixilation; the individual pixels are hard to resolve visually but when the head is moving (as it always is in VR) we can see the action along edges.   Easiest to see looking at coastlines or on runway edges although it also shows up in cockpit textures.  Certain structures -- cranes, towers, certain buildings, chimneys, posts -- are severely affected.  The NTTR map can be very distracting because of this.  Distant aircraft go through a transition as you approach.  Larger aircraft may be visible as a well defined dot up to 20 miles away.  Then the aircraft disappears, and upon further approach, re-appears as a squirming, pixelated blob that slowly resolves into a recognizable aircraft.  It is not clear to me that this degrades game play, but it can certainly be distracting.  

The solution would be to apply some sort of anti-aliasing.  Traditionally, VR has relied on supersampling, in other words, rendering an image significantly larger that what is displayed in the headset.  Crunching that image down to fit has an anti-aliasing effect in flat screen mode, and adds an extra dimension in VR.  In older, low resolution headsets, it gave the impression of looking at a high resolution world through a screen door, albeit at the expense of losing some definition in text legibility due to "squirminess."  Essentially we were trading resolution of detail for a nicer "big picture."  

MSAA is an anti-aliasing technique that focuses primarily on edges, and as a result is at least theoretically more efficient than supersampling.  Back in the day, the theory was, you could use supersampling to get the "screen door effect," then use MSAA to sharpen up and stabilize your edges, and that would give you the best image in VR. 

Unfortunately, MSAA is not particularly efficient in DCS, because of the way DCS handles rendering. Many still like the effect, and are willing to live with the performance hit.  But with modern, high resolution headsets, we seldom have the power to run both supersampling and MSAA at the same time.  For most of us, if we have the power to try anti-aliasing at all, we will have to choose one or the other. 

MSAA does a fine job of softening the image, although at great expense performance wise.  Supersampling to an extent somewhere between 100 and 150% of native resolution is not quite as good as MSAA at softening up the image, but it can take the edge off while saving a few milliseconds in GPU time.  I got the definite feeling Steam's Advanced Supersampling is doing something, at no measurable cost.

I agree with other observers, that the DCS  SSAA setting doesn't seem to be doing much in VR.  Oddly my GPU time went up a little, frames stayed the same.  Thus, possibly doing something, I'm just not sure what. 

Note, this doesn't leave a whole ton of good options for people with GPUs that can't quite drive the G2 at native resolution.  Supersampling is out, by definition.  The alternative is to dial down resolution to a point where MSAA can be applied, either by using Steam settings, or by using a scaling mod like FSR (or some combination of the two, which is an active area of research).  No guarantee this will work, as scaling introduces its own artifacts.  FSR, according to many reports, is superior to whatever protocol WMR is using, and many players are satisfied with the resulting cosmetics.  However, in my testing at least, gameplay may suffer due to significantly reduced spotting distance.  

(Note: shadows off for this test.  It's not the actual numbers that matter, it's the relative cost and benefit)

antialiasing.PNG


Edited by DeltaMike
  • Like 3

Ryzen 5600X (stock), GBX570, 32Gb RAM, AMD 6900XT (reference), G2, WInwing Orion HOTAS, T-flight rudder

Link to comment
Share on other sites

Yeah for AA MSAA in DCS is the best currently. Don't think those others do anything.

Shame they don't offer TAA. Seems to have good results without such a large performance hit.

  • Like 2

Don B

EVGA Z390 Dark MB | i9 9900k CPU @ 5.1 GHz | Gigabyte 4090 OC | 64 GB Corsair Vengeance 3200 MHz CL16 | Corsair H150i Pro Cooler |Virpil CM3 Stick w/ Alpha Prime Grip 200mm ext| Virpil CM3 Throttle | VPC Rotor TCS Base w/ Alpha-L Grip| Point Control V2|Varjo Aero|

Link to comment
Share on other sites

32 minutes ago, streakeagle said:

MSAA has always worked the best for me in terms of shimmering and reading instruments/CRTs. The problem is MSAA x2 is the most I can afford to apply.

 

Yeah x2 is what I am running and satisfied with.

Don B

EVGA Z390 Dark MB | i9 9900k CPU @ 5.1 GHz | Gigabyte 4090 OC | 64 GB Corsair Vengeance 3200 MHz CL16 | Corsair H150i Pro Cooler |Virpil CM3 Stick w/ Alpha Prime Grip 200mm ext| Virpil CM3 Throttle | VPC Rotor TCS Base w/ Alpha-L Grip| Point Control V2|Varjo Aero|

Link to comment
Share on other sites

I'm currently running steamvr 150% with vrperfkit 0.75 which theoretically equates to a equivalent rendered GPU load of about 112%. The increased resolution really crisps up the MFDs dramatically, much more readable then 100% + MSAA. I can live with the slight jaggy edges as they're really minimal in the cockpit, road edges are still a line of marching ants at certain angles though. I don't sharpen anything, anywhere as it just adds shimmer for me.

On Caucasus I can afford to turn on MSAA and it then looks glorious, frametimes are still will within the 60Hz smoothing envelope but typically dropping out of the bottom of the 90Hz smoothing envelope. On Syria I can run MSAA but it's eating into the headroom.

Ryzen7 7800X3D / RTX3080ti / 64GB DDR5 4800 / Varjo Aero / Leap Motion / Kinect Headtracking
TM 28" Warthog Deltasim Hotas / DIY Pendular Rudders / DIY Cyclic Maglock Trimmer / DIY Abris / TM TX 599 evo wheel / TM T3PA pro / DIY 7+1+Sequential Shifter / DIY Handbrake / Cobra Clubman Seat
Shoehorned into a 43" x 43" cupboard.

Link to comment
Share on other sites

^Interesting.  Although... aren't you undersampling just a tad?  I get a DCS frame size of  2800x2800, vs 3100x3100 which is what my system renders at 100/100/1.0.  No? 

That said, interesting approach.  I dunno about trying to run say 100/50/1.0 and trying to layer MSAA over that, I've tried it and it's not pretty.  But... interesting approach.  Gets you into that range where FSR can really work its magic, huh.

Have you tried something like 100/100/1.0 with a ScaleFactor of 0.9?  Do you think there's a difference?  

Ryzen 5600X (stock), GBX570, 32Gb RAM, AMD 6900XT (reference), G2, WInwing Orion HOTAS, T-flight rudder

Link to comment
Share on other sites

I think it works out as 150% image that is then downsampled by 0.75 which equates to 112.5%.  My global is 100% and per app multiplier is 150% which gives me 3868 x 3784.

The steamvr global and per app resolution multipliers aren't cumulative so 100/50 is the same as 50% (which is why it doesn't look great).  Using fsr0.75 I can run 100/100/1.0 all day long with MSAA 2x (using my maths from above - effectively running equivalent of 100/75/1.0 GPU load but with 100/100/1.0 image quality) but in my experience the sharpness of cockpit text is far improved by pushing the initial image quality up before the FSR gets it's hands on it.  This is the theory that @nikoel put forward with FSR and it's given great improvement for me.

I can read every single number on the A10C engine dials whilst sat back in the chair, obviously clarity improves as you get closer but for me it's just fine.  HUD and MFDs are crystal clear, the only caveat to that would be if the smoothing decides to artifact right on the text but that very rarely happens.

There is more edge aliasing with 150% and no MSAA than 100% and MSAA 2x but I can overlook it as the clarity is great.  MSAA works great in dcsvr for me, it really does smooth out all of the edges and nearly eliminates shimmer (there's always some cases where it doesn't).

Note that I am 100% relying on running the G2 in 60Hz mode to keep it within the smoothing envelope, I cannot guarantee to be able hold the 18ms/55fps lower threshold required to keep smoothing engaged at 90Hz.  If I had a gpu with more grunt than the 3070 then I would be running 150% with MSAA on all the time.

edit: I have tried FSR at the higher end of the scale factor but the image quality doesn't improve much more than 0.75/0.77 but performance does suffer.  I don't see any noticeable degredation of image until I get down to 0.7 or there abouts.


Edited by edmuss

Ryzen7 7800X3D / RTX3080ti / 64GB DDR5 4800 / Varjo Aero / Leap Motion / Kinect Headtracking
TM 28" Warthog Deltasim Hotas / DIY Pendular Rudders / DIY Cyclic Maglock Trimmer / DIY Abris / TM TX 599 evo wheel / TM T3PA pro / DIY 7+1+Sequential Shifter / DIY Handbrake / Cobra Clubman Seat
Shoehorned into a 43" x 43" cupboard.

Link to comment
Share on other sites

Thanks for such a systematic test @DeltaMike 

@edmussis correct. FSR mod makes the game engine render at a lower res rafter all the multiplications (PD, SVR and SVR Per App). It then upscales the image to the original final res. So lets say you are rendering a final resolution of 3500x3500px. Then you install FSR and apply 0.75 multiplier. Your graphics card renders 0.75*3500=2625px2625px -> FSR Magic™-> 3500x3500px resolution thus the improved performance*

*If you are CPU limited (like I am at times after rendering 7.5K combined resolution) then no amount of FSR will help. FSR only elevates GPU Performance bottlenecks
^ A lot of people are CPU bound. I am running a 3900X. A Gold Sample at that; a high end previous generation chip at 4.5ghz locked. I am hitting this brick wall on top of supercarrier. When in flight I am GPU bound again 

If you are rendering a resolution that is low. eg Less than what I approximate ~4k then FSR will give only mediocre improvements. YMMV

If you are using CAS instead of FSR you will get worst upscaling

If you have an AMD GPU this mod will be obsolete due to implementation of FSR via AMD Driver function. This is called RSR and will be coming soon™ (https://www.anandtech.com/show/17164/amds-fidelityfx-super-resolution-to-become-driver-feature-radeon-super-resolution)

To get good performance gains you already need a reasonable GPU/CPU combo. Unfortunately those who need it the most, will get the least out of it

I have embraced what remains of the shimmer and super high res textures and screens. It bothered me a lot in the past, but I just couldn't stomach the performance loss and clarity loss of MSAA no mater how much I wanted those dancing edges gone. Now it looks great and I no longer notice it

 

 


Edited by nikoel
  • Like 1
Link to comment
Share on other sites

12 hours ago, edmuss said:

I think it works out as 150% image that is then downsampled by 0.75 which equates to 112.5%.  My global is 100% and per app multiplier is 150% which gives me 3868 x 3784.

The size of the rendered image scales with the square of pixel density.  

Let's start with 3100x3100 per-eye, which is about what you want the DCS frame to be, pre barrel distortion.  That's approx 9.6m pixels

x 150% = 14.4m

x .77^2 = 8.6m

8.6/9.6 = 89.6%

sqrt*(8.6m) = 2923x2923

aaaaand we were both wrong.  Hopefully somebody will check our figures, otherwise I'll have to get my daughter the math major to check it, and and deal with the eye-rolling and heavy sighs.  

Quote

The steamvr global and per app resolution multipliers aren't cumulative so 100/50 is the same as 50% (which is why it doesn't look great).

New users take note: just adjust one at a time, makes the math WAY easier

Quote

Using fsr0.75 I can run 100/100/1.0 all day long with MSAA 2x (using my maths from above - effectively running equivalent of 100/75/1.0 GPU load but with 100/100/1.0 image quality) but in my experience the sharpness of cockpit text is far improved by pushing the initial image quality up before the FSR gets it's hands on it.  This is the theory that @nikoel put forward with FSR and it's given great improvement for me.

 

Question in my mind is, does pushing resolution in Steam increase the amount of information available to FSR?

To my understanding -- and I don't know, this is a total "black box approach" -- I think the system works like this

  1. SVR and WMR send a "request" to DCS to render a certain frame size (regulated by Steam resolution settings)
  2. DCS can either comply or not (regulated by PD)
  3. The resulting DCS frame is sent to the compositor for barrel distortion and scaling

So we can see where FSR does its magic -- at the level of the compositor -- but I don't see how it could improve performance without decreasing the DCS frame resolution to begin with.  In my testing, ScaleFactor has the same effect as PD on GPU time and FPS....

One way to test this might be to flip the numbers and see if it makes any difference.  Try 100/100/1.0 with a ScaleFactor of 0.95.  Or 100/90/1.0 without using FSR at all.  All three give you the same DCS frame size of about 2900x2900.  I'd bet my own money the performance would be the same; question is, whether the aesthetics would be any different. 

Regardless.  I've been kind of pushing the idea that undersampling just for the purpose of adding MSAA isn't a great idea.  I think what you've shown is, it's OK to undersample a little bit, especially if you let FSR handle the scaling, and the result is pretty darn good.

I say, bravo!!! Belissimo!!!

Quote

Note that I am 100% relying on running the G2 in 60Hz mode to keep it within the smoothing envelope, I cannot guarantee to be able hold the 18ms/55fps lower threshold required to keep smoothing engaged at 90Hz.  If I had a gpu with more grunt than the 3070 then I would be running 150% with MSAA on all the time.

New users, take note.  There are several strategies to deal with frame time spikes

  1. Reduce your settings (either resolution or graphics) to shift the curve to the left
  2. Reduce frame time variability (an option with Big Navi, unclear to me how you would do this with NVIDIA)
  3. Increase your target frame time by adjusting your refresh rate (which is like hitting the easy button)

People tend to be afraid to try the latter for fear of inducing VR sickness, but as you've shown, that's simply not the case

Also, you've done a yeoman's job tuning the 3070.  NVIDIA should sponsor you, except to the extent you're cutting into 3080 sales... eh, screw em. Those things are expensive. 

 


Edited by DeltaMike

Ryzen 5600X (stock), GBX570, 32Gb RAM, AMD 6900XT (reference), G2, WInwing Orion HOTAS, T-flight rudder

Link to comment
Share on other sites

8 hours ago, nikoel said:

Thanks for such a systematic test @DeltaMike 

@edmussis correct. FSR mod makes the game engine render at a lower res rafter all the multiplications (PD, SVR and SVR Per App). It then upscales the image to the original final res. So lets say you are rendering a final resolution of 3500x3500px. Then you install FSR and apply 0.75 multiplier. Your graphics card renders 0.75*3500=2625px2625px -> FSR Magic™-> 3500x3500px resolution thus the improved performance*

 

 

Well, the final resolution is gonna be 2100x2100.....    But, I guess the point is, if I'm understanding correctly --  if you're gonna undersample anyway, there's an argument for using the superior scaling algo, plus you get the sharpening filter besides (which in my experience helps text considerably).  

 

Quote

If you have an AMD GPU this mod will be obsolete due to implementation of FSR via AMD Driver function. This is called RSR and will be coming soon™ (https://www.anandtech.com/show/17164/amds-fidelityfx-super-resolution-to-become-driver-feature-radeon-super-resolution)

 

Interesting.  Even now, some of us have noted, something is handling the scaling, and it's not doing a half bad job.  100/90/1.0 really doesn't look bad at all.  

  • Like 1

Ryzen 5600X (stock), GBX570, 32Gb RAM, AMD 6900XT (reference), G2, WInwing Orion HOTAS, T-flight rudder

Link to comment
Share on other sites

Admittedly I hadn't trawled through the maths and I have probably massively oversimplified it from my point of view, one other factor that we don't know is how the 0.75 FSR renderscale is applied to the image, is it 0.75 of each value or 0.75 of the total image?

If it takes 75% of the X and Y resolution and then upscales then yes it would be giving a final render of 8.4 million pixels, however if it takes 75% of the total number of rendered pixels then it would be giving a final render of 10.9 million pixels.  Whichever it uses is a moot point as the scale just sidesteps one way or the other.

In addition to vrperfkit the only other GPU affecting mods are kegetys shader mod (gives a couple of free ms frametime) and I've undervolted the 3070 so it stays cooler.  Amusingly if smoothing is engaged, it drops the GPU load to around 55% and hence the temperature plummets, I often have it sat at < 60°C when flying.

In my experience bumping up the steamvr resolution didn't actually cost that much more frametime, conversely dropping the resolution didn't give that much reduction in frametime.  The inital cost of entry into DCSVR is high from the point of GPU performance, but once you're in and stable I found that I've been able to turn the settings up really quite high for a mid range rig.  I have made a concession to run low gound textures and flat terrain shadows, but the rest is all currently on high.

Ryzen7 7800X3D / RTX3080ti / 64GB DDR5 4800 / Varjo Aero / Leap Motion / Kinect Headtracking
TM 28" Warthog Deltasim Hotas / DIY Pendular Rudders / DIY Cyclic Maglock Trimmer / DIY Abris / TM TX 599 evo wheel / TM T3PA pro / DIY 7+1+Sequential Shifter / DIY Handbrake / Cobra Clubman Seat
Shoehorned into a 43" x 43" cupboard.

Link to comment
Share on other sites

Boys I’m just going to leave this here. Copy and paste from the FSR Mod itself on how the multiplier works

Also this presentation on how Steam VR works (by valve themselves) is possibly the best in explaining some of the reasoning behind how things are and why they were implemented in such a way.  Antialiasing and resolution I believe is mid to end of the presentation 

 

  # method to use for upscaling. Available options (all of them work on all GPUs):
  # - fsr (AMD FidelityFX Super Resolution)
  # - nis (NVIDIA Image Scaling)
  # - cas (AMD FidelityFX Contrast Adaptive Sharpening)
  method: fsr
  # control how much the render resolution is lowered. The renderScale factor is applied to both width and height. So if renderScale is set to 0.5 and you have a resolution of 2000x2000 configured in SteamVR, the resulting render
 resolution is 1000x1000.
 NOTE: this is different from how render scale works in SteamVR! A SteamVR render scale of 0.5 would be equivalent to renderScale 0.707 in this mod!


Edited by nikoel
  • Like 1
Link to comment
Share on other sites

Thanks a lot for the testing!

Personally I don't use MSAA because it is hacked into DCS right now. When DCS switched to deferred rendering a new AA solution was required because MSAA doesn't work very well in deferred rendering and it impacts the performance a lot, but we never got it.

One of the advantages of deferred rendering is to have multiple light sources without extra performance impact, ever wondered why ED introduced deferred rendering but we never got real lighting from light sources in DCS? Yep, you guessed right, because thanks to the ancient MSAA the performance will be destroyed (even more).

On VR the issues are much more noticeable and MSAA with deferred rendering creates the infamous 'shimmering', which we will not get rid of no matter how much MSAA or supersampling we apply with our NASA PC. You can turn off MSAA but then you have a non-aliased image with shimmering as well. There is a reason why optimised VR games like Alyx use forward rendering with MSAA. We need alternative AA solution for deferred rendering like TAA or DLSS, that will solve many many of our VR picture quality issues in DCS. Hopefully with Vulkan.

  • Like 3

i7 12700KF | MSI Z690 A-PRO | Corsair Vengeance 2x16 gb @ 3200 Mhz | RTX 3070 Ti FE | Acer XB271HU 1440P 144HZ | Virpil T-50 CM throttle | Virpil WarBRD Base + MongoosT-50 CM2 Grip | MFG Crosswind | TrackIR 5 | HP Reverb G2

Bf 109 K-4 | Fw 190 A-8 | Spitfire LF Mk. IX | P-51D | Fw 190 D-9 | P-47D | Mosquito FB VI | F/A 18C | F-14 A/B | F-16C | MiG-15bis | MiG-21bis | M-2000C | A-10C | AJS-37 Viggen | UH-1H | Ka-50 | Mi-24P | C-101 | Flaming Cliffs 3

Persian Gulf | Nevada | Normandy | The Channel | Syria

Link to comment
Share on other sites

I believe that vrperfkit implements either fsr or nis independantly of what GPU you have, I haven't used nis recently but I have previously tried it with very similar results to fsr.  Note that I haven't tried running nis with the 150% steamvr resolution so can't comment on how it looks.

  • Like 1

Ryzen7 7800X3D / RTX3080ti / 64GB DDR5 4800 / Varjo Aero / Leap Motion / Kinect Headtracking
TM 28" Warthog Deltasim Hotas / DIY Pendular Rudders / DIY Cyclic Maglock Trimmer / DIY Abris / TM TX 599 evo wheel / TM T3PA pro / DIY 7+1+Sequential Shifter / DIY Handbrake / Cobra Clubman Seat
Shoehorned into a 43" x 43" cupboard.

Link to comment
Share on other sites

So, what if we can go outside DCS, and get a post-processing filter?  Here, I test ReShade 5.0.2.  G2, 6900xt, 100/100/1.0.  

Observations

  1. FXAA and SMAA have a similar effect on frame time and FPS.  While FPS is grossly about the same as with MSAA, I feel there's a significant improvement in frame time.  About two and a half milliseconds, which is something I suppose.   
  2. Both have a noticeable effect on shimmering.  Right out of the box, FXAA results in much better definition of "chunky" objects like buildings and bridges, also quite a bit less shimmering along runway edges.  SMAA maybe not quite as good.  Neither one makes power lines look worth a darn.  

Both of these settings are adjustable, particularly SMAA.   Tweaking those settings can result in a pretty darn good image.  And of course you get the other ReShade effects.  I've been flying around with Curves, Lumasharpen and SMAA this morning and it looks great actually.  Even on NTTR.  The apron still looks like you've taken just a little too much acid to be flying.  But, flying over the city, it looks... well,  like a brilliant sunny day in the desert.  MFD's look quite a bit nicer, but you know how much I like sharpening filters, YMMV.  

If you want to try this, a couple of hints

  • The only way to affect the VR image is to select and adjust your effects from within VR.
  • You do this by bringing up your Steam Dashboard.  ReShade helpfully puts an icon on your toolbar.
  • Check "performance mode" to select and apply effects.
  • Uncheck "performance mode" to tweak the effect
  • There is no need for hotkeys, which is good, because they don't work anyway.  Or more specifically, the hotkeys only work on the mirror.  Which is kind of cool in itself, you can separately adjust the appearance of the mirror, from what you see in VR.  Useful I suppose if you're shooting video, I guess?

Keep in mind your target framerate.  You kind of want to keep that below 20ms if you can.  That makes it practical to use motion reprojection at 90hz.  Also it keeps you from puking.  Lot easier to hit that target with post-processing than with MSAA, and results are better than "splitting the difference" with supersampling.  Not bad for a cost of one millisecond basically.  Not bad at all. 

 

rshade.PNG


Edited by DeltaMike

Ryzen 5600X (stock), GBX570, 32Gb RAM, AMD 6900XT (reference), G2, WInwing Orion HOTAS, T-flight rudder

Link to comment
Share on other sites

It hurts me to say this, but all of our testing might have just been undone

I have just tried the OpenXR hack... mod... thing... It has given me a ridiculous visual and performance uplift. FSR sadly doesnt work. It does not matter. Reshade doesnt work either. To me it still looks better. I don't know how many FPS I am getting. Frame time improvements - no idea (FPSVR doesn't work). But the perceived difference is significant. Texture quality improvement alone is mind blowing. All I did was set a PD of 1.1 in the DCS settings to bring performance back and visual quality up and went straight to OpenXR settings and unticked motion smoothening as is the way. I didn't want to mess with the opencomposite.ini and left it as default. All AA settings to off

And there is no more epilepsy inducing BRRRRT loading screens

https://gitlab.com/Jabbah/open-composite-acc

Make sure to install the d3dcompiler

Make sure to open Mixed Reality Portal and quit/don't open SteamVR and keep the DCS window on top during the boot of the game

Before I installed the mod, I ran a full clean and repair

F-14 is Crisp. Feels like I have upgraded to a newer generation headset

I will leave it at that


Edited by nikoel
Link to comment
Share on other sites

It hurts me to say this, but all of our testing might have just been undone
I have just tried the OpenXR hack... mod... thing... It has given me a ridiculous visual and performance uplift. FSR sadly doesnt work. It does not matter. Reshade doesnt work either. To me it still looks better. I don't know how many FPS I am getting. Frame time improvements - no idea (FPSVR doesn't work). But the perceived difference is significant. Texture quality improvement alone is mind blowing. All I did was set a PD of 1.1 in the DCS settings to bring performance back and visual quality up and went straight to OpenXR settings and unticked motion smoothening as is the way. I didn't want to mess with the opencomposite.ini and left it as default. All AA settings to off
And there is no more epilepsy inducing BRRRRT loading screens
https://gitlab.com/Jabbah/open-composite-acc
Make sure to install the d3dcompiler
Make sure to open Mixed Reality Portal and quit/don't open SteamVR and keep the DCS window on top during the boot of the game
Before I installed the mod, I ran a full clean and repair
F-14 is Crisp. Feels like I have upgraded to a newer generation headset
I will leave it at that
For G2 it didn't work, right?

Inviato dal mio BLA-L09 utilizzando Tapatalk

Link to comment
Share on other sites

I am running 3900X, 6900XT, with a Reverb G2

Yeah it works alright. As long as I keep clicking on the DCS icon on the taskbar during boot. Otherwise black screen

So, I have discovered OpenXR Toolkit and our ticket to getting FSR back. Installed it and tried to use it. The log files show that the window is activating (via CTRL+F2) and everything is running perfectly. Except no darn window is coming up

Ugh... Help? https://mbucchia.github.io/OpenXR-Toolkit/

Link to comment
Share on other sites

Today I ran through some various combinations of ReShade effects.  NTTR map, on the ramp at Nellis.  100/100/1.0.

Reminder: point of this thread is, let's say you can run your VR headset at a reasonable approximation of native resolution with a few milliseconds of GPU time to spare.  What's the best use of that time?  Today I go through some of your post-processing options for anti-aliasing, available through ReShade. 

Observations

Baseline:  heh I need some work on this map.  FPS all over the place.  Main thing, as usual, is to look at the GPU time delta.

SMAA: Most report it costs about 1.5ms.  I think it's doing something.

SMAA + Curves + Lumasharpen:  Looks awesome, gets me right to my personal threshold of 20ms (frames improve once you're in the air). Note how sharp the HUD is. I know that has nothing to do with AA but still. LumaSharpen is pretty awesome. 

HQAA: Applies at least two passes with SMAA and one of CAS with really good effect.  Costs me 23% GPU-time-wise, I'd be better off with MSAA I think.  Plus MSAA gives a softer image, which may be more what you're looking for.  

TAA + SMAA:  Excellent effects on shimmering, as long as you hold your head still (some advocate it for flat screen but I don't see it for VR)

SMAA + NFAA: A better anti-shimmering solution

DLAA: if you must get rid of shimmering at all costs, this should be on your short list.  Expensive, but it works

VR Toolkit: as advertised, you get FXAA, CAS and color/contrast correction, all in one pass.  As advertised, this is super efficient and deserves a close look imo.  Note: if you're trying to reduce shimmering, go easy on the sharpening filters.  Very easy to undo what your AA did for you.  

Pics:

eUV9BpH.jpg Baseline qbr94rr.jpg SMAA + Curves + Lumasharpen

3PpIONp.jpg HQAA 3PpIONp.jpg VR Toolkit

 

 

 

reshade_aa2.PNG


Edited by DeltaMike

Ryzen 5600X (stock), GBX570, 32Gb RAM, AMD 6900XT (reference), G2, WInwing Orion HOTAS, T-flight rudder

Link to comment
Share on other sites

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...