DLSS & VR

[Lorenzo.D]

Tested for hours and hours. 
I have had the absolute best results running with:

• DLAA
• 72Hz (no reprojection!)
• 3500x3500 resolution in Steam VR 
• Quad FOV rendering (latest version)
• Balanced settings in DCS to get stable 72 fps

Rig details in signature…
 

What do quad fov rendering? I don t know It. I do not have Eye tracking on quest' 3

Inviato dal mio CPH2219 utilizzando Tapatalk

[Koekemoeroetoe]

Unfortunately, as already expected, for me it looks like DLSS is no good. I think you can only use it to make already high FPS even higher, but since I'm starting with low FPS the overall experience only decreases with DLSS enabled.

On my RTX 2070/i9 10850k/64gb RAM and Reverb G2 at 100% resolution, I get about 30-40 FPS. Enabling DLSS quality + DLAA (which is forced on for me) gives the same FPS with less shimmering, but reduced clarity and some ghosting. Less quality in general. Any DLSS performance mode only decreases clarity further and massively increases ghosting, up to the point where MFD's are not readable anymore.

Time to upgrade the GPU I guess...

[MIghtymoo]

1 hour ago, Lorenzo.D said:

What do quad fov rendering? I don t know It. I do not have Eye tracking on quest' 3

Inviato dal mio CPH2219 utilizzando Tapatalk
 

I don’t have eye tracking either on my Pico 4. Still works grest and gives improved performance/fps.

In each chart, the top bar corresponds to the use of quad views with foveated rendering, while the bottom graph corresponds to stereo views without foveated rendering.

—————

Super quick and easy to setup for my headset.

More info:

https://github.com/mbucchia/Quad-Views-Foveated/wiki/What-is-Quad-Views-rendering%3F

 

Edited October 21, 2023 by MIghtymoo

[Qcumber]

16 minutes ago, MIghtymoo said:

I don’t have eye tracking either on my Pico 4. Still works grest and gives improved performance/fps.

In each chart, the top bar corresponds to the use of quad views with foveated rendering, while the bottom graph corresponds to stereo views without foveated rendering.

—————

Super quick and easy to setup for my headset.

More info:

https://github.com/mbucchia/Quad-Views-Foveated/wiki/What-is-Quad-Views-rendering%3F

 

 

I am getting a similar improvement using QVFR.

 

36 minutes ago, Koekemoeroetoe said:

Unfortunately, as already expected, for me it looks like DLSS is no good. I think you can only use it to make already high FPS even higher, but since I'm starting with low FPS the overall experience only decreases with DLSS enabled.

On my RTX 2070/i9 10850k/64gb RAM and Reverb G2 at 100% resolution, I get about 30-40 FPS. Enabling DLSS quality + DLAA (which is forced on for me) gives the same FPS with less shimmering, but reduced clarity and some ghosting. Less quality in general. Any DLSS performance mode only decreases clarity further and massively increases ghosting, up to the point where MFD's are not readable anymore.

Time to upgrade the GPU I guess...

Have you tried QVFR? I have used this to push the centre resolution to 1.5 and it gives me a clearer image with DLSS and less ghosting. 

[zildac]

I'm seeing a ghosting/rippling effect when on take off roll around hornet canopy when using TAA. Really odd, shame as TAA looks better to me than DLAA (taxiway lines etc).

 

[MIghtymoo]

15 minutes ago, zildac said:

I'm seeing a ghosting/rippling effect when on take off roll around hornet canopy when using TAA. Really odd, shame as TAA looks better to me than DLAA (taxiway lines etc).

 

Same experience here.

[zildac]

Same experience here.

Thanks for confirming. Not sure if it's just TAA or the DCS implementation.

[Lorenzo.D]

Same experience here.

I don t understand Is Better taa or dlaa?

Inviato dal mio CPH2219 utilizzando Tapatalk

[Lorenzo.D]

I don’t have eye tracking either on my Pico 4. Still works grest and gives improved performance/fps.


In each chart, the top bar corresponds to the use of quad views with foveated rendering, while the bottom graph corresponds to stereo views without foveated rendering.

—————

Super quick and easy to setup for my headset.
More info:
https://github.com/mbucchia/Quad-Views-Foveated/wiki/What-is-Quad-Views-rendering%3F
 

But the pancake lenses are good for see all the area of the lenses...so with It you have less clarity on the Edge likes you don t have a pancake ....

Inviato dal mio CPH2219 utilizzando Tapatalk

[MIghtymoo]

28 minutes ago, Lorenzo.D said:

But the pancake lenses are good for see all the area of the lenses...so with It you have less clarity on the Edge likes you don t have a pancake ....

Inviato dal mio CPH2219 utilizzando Tapatalk
 

No issue whatsoever. Quadview FOV is so great that I had to doublecheck to see if it was activated or not first time I used it.

OXRTK FOV rendering  is much more visible.

[Hoirtel]

Is quad views actually better performing than the normal foveated rendering? Both without an eye tracked headset. 

[zildac]

I don t understand Is Better taa or dlaa?

Inviato dal mio CPH2219 utilizzando Tapatalk

For me TAA offers the best visual quality. Especially when looking at taxi way lines and the like.

[Hippo]

DLAA + DFR (i.e. mbucchia quad views foveated) is working wonders for me on Quest Pro.  DLAA appears to completely elimate shimmering in the peripheral area meaning the focus region can be substantially reduced for huge performance improvements and a much better overall experience.  This is DFR working as I always imagined.

 

23 minutes ago, zildac said:

1 hour ago, Lorenzo.D said:

I don t understand Is Better taa or dlaa?

Inviato dal mio CPH2219 utilizzando Tapatalk
 

For me TAA offers the best visual quality. Especially when looking at taxi way lines and the like.

Whilst TAA greatly reduces shimmering,  I find it just makes everything blurry, and can produce dreadful ghosting (e.g. in F-18 try quickly looking left and right past canopy frame).  DLAA much better but much greater perf hit, same as MSAAx4.  But see above if you're using DFR.

Having said that, I am running a pretty high-end system.  For something more mid-range TAA could be a godsend, it greatly improves the terrain by eliminating shimmering at almost no perf cost, and I think that the terrain blurring looks ok.  However, I don't like blurring of the cockpit.  Swings and roundabouts.

Edited October 21, 2023 by Hippo

[slughead]

11 hours ago, twistking said:

Those that have bad ghosting with DLSS, try to run without ASW/Reprojection, even if that means you have to set a more aggressive DLSS setting. DLSS will never look good with ASW/Reprojection, because it has less temporal information to work with, while at the same time the resulting artifacts (smearing/ghosting) will stay visible longer.

If you run a quest, try upping from 72 to 90 hz. DLSS will look cleaner and higher resolution with higher refresh rates, as it gets more temporal information to work with.

Do not run DLSS with low framerates. More fps make better visuals.

With a risk of sounding like Alec Guiness in Star Wars, "These are not the ghosts you are thinking of". The ghosting (double images) of fast-moving objects with reprojection/ASW is not the same as the effects people are referring to with DLSS/DLAA. I prefer to call this "smearing". It's akin to a firey phoenix with flames or smoke trailing off the edges of wings and fuselage. It occurs with our without reprojection/ASW and at all refresh rates and render resolutions.

[Lorenzo.D]

Has anyone experienced continuous shaking when switching to the new DCS 2.9? It feels like my head is shaking but it's not

Quest3 virtual desktop steamvr openxr

Inviato dal mio CPH2219 utilizzando Tapatalk

[Swamp Fox Loader]

48 minutes ago, zildac said:

1 hour ago, Lorenzo.D said:

I don t understand Is Better taa or dlaa?

Sent from my CPH2219 using Tapatalk
 

For me TAA offers the best visual quality. Especially when looking at taxi way lines and the like.

I agree zildac.  TAA on my system looks the best.  I also turn sharpening up to 1 in DCS settings, and sharpening on OpenXR Toolkit to 100.

[Motomouse]

TAA looks awful at 45FPS. Parts of the other aircrafts disappearing and reappearing all the time. Spotting is bad.

DLAA for me all the way.

[zildac]

Interesting to hear such differing views. DLAA is excellent, but I still see the shimmers on taxi way markings. And as a few people have mentioned, TAA does cause some pretty bad ghosting when moving fast. I can't say I noticed any blurring as a result of TAA though. I think I need to do some more testing!  Specs in sig.

52 minutes ago, slughead said:

With a risk of sounding like Alec Guiness in Star Wars, "These are not the ghosts you are thinking of". The ghosting (double images) of fast-moving objects with reprojection/ASW is not the same as the effects people are referring to with DLSS/DLAA. I prefer to call this "smearing". It's akin to a firey phoenix with flames or smoke trailing off the edges of wings and fuselage. It occurs with our without reprojection/ASW and at all refresh rates and render resolutions.

Agreed, accurate description of the phenomenon

[Hippo]

48 minutes ago, Lorenzo.D said:

Has anyone experienced continuous shaking when switching to the new DCS 2.9? It feels like my head is shaking but it's not

Quest3 virtual desktop steamvr openxr

Inviato dal mio CPH2219 utilizzando Tapatalk
 

I've had no issues like that.  But then I'm running a Quest Pro, the stanalone version of DCS (i.e. not Steam), and am using a link cable with the Oculus software (i.e. not VD).

[Chapa]

DLAA has changed everything for me. Now I can use mbucchia quad views foveated because the peripheral view doesn't shimmer with DLAA.

I have been able to reduce the resolution in the oculus app from x1.5 to x1.2 and add terrain shadows and grass.

Really happy with 2.9. Good job ED and thank you mbucchia for your work.

 

 

 

Edited October 21, 2023 by Chapa

[Hippo]

2 minutes ago, Chapa said:

DLAA has changed everything for me. Now I can use mbucchia quad views foveated because the peripheral views doesn't shimmer with DLAA.

I have been able to reduce the resolution in the oculus app from x1.5 to x1.2 and add terrain shadows and grass.

Really happy with 2.9. Good job ED and thank you mbucchia for your work

Agreed.  After further testing, I feel this is truly a game changer.  By setting the focus region at 0.25 I am getting AA at a quality very similar to MSAAx4 for free.  Yet it is almost impossible for me to tell that DFR is doing its thing (I can see the "square" if I look for it and depending on what I'm looking at, but it's something I can easily live with).  I'm possibly seeing some texture DLAA shimmering in the focus region that wasn't there with MSAA (further testing required to confirm). but again, minor enough that I can easily live with it.

P.S. Night vision goggles now work properly, and it is possible to see the cockpit "outside" of them.  Xmas has come early.  Thanks to all concerned.

[twistking]

16 hours ago, twistking said:

Those that have bad ghosting with DLSS, try to run without ASW/Reprojection, even if that means you have to set a more aggressive DLSS setting. DLSS will never look good with ASW/Reprojection, because it has less temporal information to work with, while at the same time the resulting artifacts (smearing/ghosting) will stay visible longer.

If you run a quest, try upping from 72 to 90 hz. DLSS will look cleaner and higher resolution with higher refresh rates, as it gets more temporal information to work with.

Do not run DLSS with low framerates. More fps make better visuals.

4 hours ago, slughead said:

With a risk of sounding like Alec Guiness in Star Wars, "These are not the ghosts you are thinking of". The ghosting (double images) of fast-moving objects with reprojection/ASW is not the same as the effects people are referring to with DLSS/DLAA. I prefer to call this "smearing". It's akin to a firey phoenix with flames or smoke trailing off the edges of wings and fuselage. It occurs with our without reprojection/ASW and at all refresh rates and render resolutions.

Missing my point you do, young padawan!

DLSS creates a high resolution image (and is technically able to create perceived higher res than native in scenes with predictable or constantly slow movment). This is done by subpixel jittering an temporal accumulation of information, so the image gets refined "over time" - or more precisely over a sequence of sequential frames. Therefore in the same scene with moderate movment, double fps should mean more clarity, because a good estiamtion can be made in half the time. Additionally on higher fps, changes in the scenes are more gradual, so the guessing is easy and the algorythm should have an easier time "understanding" motion of the scene. Motion vectors only help little, when you have massive difference between two frames, because of fast relative motion. Also whenever the algorythm does a happy accident (which happens everywhere, all the time... to some degree) the resulting artifacts should be smaller when relative motion is slower, or when the difference between sequential images is lower, which is the case - again - on higher fps.

Now imagine what happen if you have low FPS - so more guesswork for DLSS and potentially more visible artifacts to begin with - and now introduce another algorythm that seperately from DLSS interpolates between sequential frames to create a synthetic frame from two sequential frames that are already potentially impaired by motion vector guesswork. It's an unholy alliance of two "AI" systems playing a game of telephone.

This is all a purely theoretical take on DLSS. I can't try it for myself unfortunately. I'm also not implying that your observations and personal experiences are wrong. Maybe the effects i describe are not that relevant in practice, but they must be technically true. Maybe keep an open eye for those aspects and keep an open mind when doing testing and tweaking. FPS must have an effect on the quality of DLSS (maybe not a relevant one though). I still think it's worth keeping in mind.

I also strongly suspect that users reporting very bad image quality with DLSS do not actually stumbled on the limitations of DLSS, but encountered implementation bugs, run bad settings or similar. Also some issues reported would clearly be solved by doubling FPS: For example the mechanical digits in analogie gauges being blurry when changing the displayed numbers. On 45fps there is simply not enough temporal resolution to resolve that motion - or for the "AI" to "udnerstand" it.

Again. I'm not saying that DLSS is good and your observations are bad. But completely denying the effects of FPS is not helpful for troubleshooting and understanding artifacting - even if the actual differences between low and high FPS are mabye indeed smaller than i would expect from a purely theoretical standpoint. Hope that makes sense.

Thanks!

Edited October 21, 2023 by twistking

[slughead]

10 minutes ago, twistking said:

This is all a purely theoretical take on DLSS. I can't try it for myself unfortunately

Thanks for your insights. However, I have actually tested this. The results are the same in 72fps or 90 with or without ASW.

 

11 minutes ago, twistking said:

I also strongly suspect that users reporting very bad image quality with DLSS do not actually stumbled on the limitations of DLSS, but encountered implementation bugs, run bad settings or similar

This is because people have jumped into the cockpit with the same Oculus / supersampling settings for their headset as they had before 2.9. Once they realise they need to increase their render resolution to get the clarity back, they'll stop moaning as I did. Those that are saying it's great are probably already running at a high enough render resolution.

 

May the force be with you.

[Qcumber]

14 hours ago, slughead said:

Thanks for your insights. However, I have actually tested this. The results are the same in 72fps or 90 with or without ASW.

 

This is because people have jumped into the cockpit with the same Oculus / supersampling settings for their headset as they had before 2.9. Once they realise they need to increase their render resolution to get the clarity back, they'll stop moaning as I did. Those that are saying it's great are probably already running at a high enough render resolution.

 

May the force be with you.

I would agree with this. I have used QVFR to increase the centre resolution by 1.8 and this produces good results with no smearing. It produces better FPS than native resolution settings and DLAA. 

Edit: after more testing, I am noticing blurring behind fast moving planes etc. but only when they are small. This appears in both DLSS and DLAA.

Edited October 22, 2023 by Qcumber

[twistking]

1 hour ago, slughead said:

Thanks for your insights. However, I have actually tested this. The results are the same in 72fps or 90 with or without ASW.

This is because people have jumped into the cockpit with the same Oculus / supersampling settings for their headset as they had before 2.9. Once they realise they need to increase their render resolution to get the clarity back, they'll stop moaning as I did. Those that are saying it's great are probably already running at a high enough render resolution.

May the force be with you.

I see. If you say that 72fps with ASW (so basically DLSS working with 36fps) does not make artifacting notably worse than true 90, then this means that my point holds no value in practice. That mean that those artifacts come from lack of spatial information instead of temporal. Not very intuitive, but i have no reason to doubt you.

I'm still quite confident that low FPS will introduce additional artifacting on image details that change details faster than can temporally be resolved on lower FPS, meaning no effective motion vector can be build from it. F.e. mechnical rotary counters changing digits... In these situations it seems logical that DLSS could even make details worse compared to the (low) base resolution without upscaling.

39 minutes ago, Qcumber said:

I would agree with this. I have used QVFR to increase the centre resolution by 1.8 and this produces good results with no smearing. It produces better FPS than native resolution settings and DLAA. 

I would be interested in seeing how DLAA and DLSS compare visually, if you set your high pixel density for use with DLSS and compare to DLAA not with native res, but with a resolution that gives similar fps to the DLSS setup. Or in other words: DLSS with high pixel vs DLAA with pixel density slightly higher than native to matching fps.

Edited October 21, 2023 by twistking