DLSS in DCS

[mbucchia]

3 minutes ago, j9murphy said:

@mbucchia  Does this tool give us any uplift once DCS supports DLSS (assuming the rumor is true that DLSS is being implemented in DCS)?  The tool essentially allows any DLSS game to use DLAA.

https://hothardware.com/news/plugin-enable-dlaa-in-dlss-games

 

 

I know nothing about that tool, though looking at their code very briefly, I don't see a reason why it wouldn't work. It's a DLL injection, so sometimes there are freaky edge cases that can make it not work (like forgetting to passthrough a symbol), but it looks pretty solid or at least fixable if needed.

[Supmua]

DLAA with DCS will probably yield the best image fidelity, although I’m probably going to use DLSS mostly since its upscaling is good enough and the performance gain is substantial in most games.  May be ED will also think about adding some raytracing options to the future update, would be awesome to see RT shadows and reflections in the cockpits.

Edited February 16, 2023 by Supmua

[Stefem]

On 2/14/2023 at 10:18 PM, Rifter said:

VR motion reprojection is a ‘forward looking’ motion estimation method. It needs the video encoder block of the GPU to do so.  Since motion reprojection is done asynchronously (in parallel to the application) it has no impact on CPU or GPU tasks. Hence the Oculus/Meta expression 'Asynchronous Space Warp'. It is until today the best method of creating additional frames for VR with repect to latency.

DLSS3 is a ‘backwards looking’ interpolation method which creates significantly increased latency. I lack the imagination how that could be compensated. Spatial reprojection is for sure no solution because that won’t work for longer time periods. DLSS3 is only for pancake.

That's the point, they have a different scope and NVIDIA chosen interpolation over extrapolation because it offered the best quality, depending on the game but DLSS frame generation doesn't add so much latency, sometimes is even lower than on AMD cards or before the development of Reflex

[Stefem]

18 hours ago, Supmua said:

DLAA with DCS will probably yield the best image fidelity, although I’m probably going to use DLSS mostly since its upscaling is good enough and the performance gain is substantial in most games.  May be ED will also think about adding some raytracing options to the future update, would be awesome to see RT shadows and reflections in the cockpits.

 

RT shadows looks really spectacular for highly detailed models, RT AO would also be great for external camera view and ED made a lot of efforts to make glass look good with rasterization which is not that easy but with ray tracing they can do a lot of nice things, even outside of graphics it can be use to simulate radars

[Supmua]

6 hours ago, Stefem said:

RT shadows looks really spectacular for highly detailed models, RT AO would also be great for external camera view and ED made a lot of efforts to make glass look good with rasterization which is not that easy but with ray tracing they can do a lot of nice things, even outside of graphics it can be use to simulate radars

HDR support would be great also (of course we still don’t have HDR capable hmd).  I used to be skeptical but newer games such as Hogwarts Legacy in HDR and full RT mode really floored me.  Spent hours calibrating my OLED TV settings just for that game.

Edited February 17, 2023 by Supmua

[Elliot]

On 1/7/2023 at 6:11 PM, SharpeXB said:

There would be no point to additional frame generation in VR via the GPU. That’s what ASW etc does. DLSS 3.0 is rather like that but for 2D

The problem with ASW though is that it looks garbage, you get a ton of artifacts for barely any extra performance. 

A good example is when you look at a sidewinder on the wing of the F-18 and begin rolling; it appears like the sidewinder has some weird motion blur. 

DLSS 3 would be better, it has dedicated hardware instead of software-only frame generation. 

 

Edited February 19, 2023 by Elliot

[MHAce75]

vor 45 Minuten schrieb Elliot:

The problem with ASW though is that it looks garbage, you get a ton of artifacts for barely any extra performance. 

A good example is when you look at a sidewinder on the wing of the F-18 and begin rolling; it appears like the sidewinder has some weird motion blur. 

DLSS 3 would be better, it has dedicated hardware instead of software-only frame generation. 

 

 

My understanding is that the aggravated artefacting is a current known bug and being worked on, though.

[some1]

1 hour ago, Elliot said:

The problem with ASW though is that it looks garbage, you get a ton of artifacts for barely any extra performance. 

A good example is when you look at a sidewinder on the wing of the F-18 and begin rolling; it appears like the sidewinder has some weird motion blur. 

DLSS 3 would be better, it has dedicated hardware instead of software-only frame generation. 

ASW doubles the visible framerate, just like DLSS3 Frame Generation does. The artifacts in ASW come from the fact that the additional frames are extrapolated from the past frames, not interpolated from the frames that have already been created by the game engine, like is the case with DLSS3. It's hard to say if DLSS would produce better quality images than ASW if it was asked to do the same thing. Even when interpolating, DLSS3 still produces visible artifacts in some situations, just like ASW.

[SharpeXB]

2 hours ago, Elliot said:

DLSS 3 would be better

That’s overly optimistic. I’m sure it’s much more difficult to double frames in 3D 6DOF than to simply generate them in 2D. Either way there are bound to be artifacts. 

[Hairdo1-1]

So I’ve tried out DLSS in VR in MSFS and while the performance is great, the displays become almost unreadable. Not sure if this will be an issue with DCS though 

[Stefem]

On 2/19/2023 at 3:42 PM, MHAce75 said:

My understanding is that the aggravated artefacting is a current known bug and being worked on, though.

It's not a bug, because of how it works and its restrain it's inherently subject to those artifact, things can be better or worst but the quality degradation is pretty evident

On 2/19/2023 at 4:23 PM, some1 said:

ASW doubles the visible framerate, just like DLSS3 Frame Generation does. The artifacts in ASW come from the fact that the additional frames are extrapolated from the past frames, not interpolated from the frames that have already been created by the game engine, like is the case with DLSS3. It's hard to say if DLSS would produce better quality images than ASW if it was asked to do the same thing. Even when interpolating, DLSS3 still produces visible artifacts in some situations, just like ASW.

That's not quite right, unlike DLSS that is interleaving a generated frame every rendered one VR reprojection is designed to generate a frame only if the next is going to miss the frametime target, the quality of the generated frames is substantially lower in contrast to DLSS ones that tend to be indistinguishable to the point that you can extract only the generated frame from a recording and it may still hard to spot any while reprojected frame quality is acceptable only because even if they look crap it's still better than missing a frame.

On 2/19/2023 at 5:51 PM, SharpeXB said:

That’s overly optimistic. I’m sure it’s much more difficult to double frames in 3D 6DOF than to simply generate them in 2D. Either way there are bound to be artifacts. 

It's not more complicated it just have more works as require two instances, one for each eye, but frame generation in its current form isn't designed for VR, they can develop a specific mode for it and VR systems dev can themselves make use of the updated OFA present on the RTX 40 series card to dramatically improve quality

[some1]

59 minutes ago, Stefem said:

unlike DLSS that is interleaving a generated frame every rendered one VR reprojection is designed to generate a frame only if the next is going to miss the frametime target

Unless you have a very fluctuating framerate that turns reprojection on/off, the ratio is the same. You end up with one "real" frame interleaved with one "fake". Or two fakes if you support reprojection at 1/3rd of headset refresh rate. The frames are generated differently with each method, but I've already said that before.

[SharpeXB]

2 hours ago, Stefem said:

It's not more complicated it just have more works as require two instances, one for each eye, but frame generation in its current form isn't designed for VR, they can develop a specific mode for it and VR systems dev can themselves make use of the updated OFA present on the RTX 40 series card to dramatically improve quality

Well clearly there’s more to it than that. If simply that the faux frames don’t match up for each eye hence the artifacts. If frame generation was flawless and easy nobody would be selling expensive graphics cards anymore. This sort of thing is bound to have issues or at least a quality trade off. It’s not a magic bullet. If DLSS 3.0 tried to work in VR it would end up with the same issues as ASW trying to do the same job. 

[Stefem]

1 hour ago, some1 said:

Unless you have a very fluctuating framerate that turns reprojection on/off, the ratio is the same. You end up with one "real" frame interleaved with one "fake". Or two fakes if you support reprojection at 1/3rd of headset refresh rate. The frames are generated differently with each method, but I've already said that before.

This drawing is just an example to explain how it works as that's actually what happen when the frametime target dictated by 90Hz V-Sync is going to be missed even once, the reprojection system will send a generated frame because the rendered one will not be ready on time (there is a system that monitor this) and the app have to wait an additional 11ms to send the frame which will result in 45fps as long as the app miss the 11ms frametime target.

Unlike DLSS frame generation that it is designed to work exclusively interleaving a generated frame after each rendered one this isn't a desirable condition for reprojection as the quality degradation is too striking, it's pretty noticeable even with just few generated frames here and there but of course is better than a stutter in VR. DLSS doesn't have this problem as the quality is impressively solid, I've attached a video made extracting only the generated frames from a recording, do you think reprojection come anywhere near this?

[mbucchia]

2 hours ago, Stefem said:

I've attached a video made extracting only the generated frames from a recording, do you think reprojection come anywhere near this?

I don't think anybody is arguing that the image quality of DLSS Frame Generation is superior to Motion Reprojection. The issue is whether DLSS Frame Generation is applicable as a replacement of Motion Reprojection in VR.

DLSS Frame Generation uses interpolation between 2 already-rendered frames (also known as backward reprojection), which is how such great quality can be achieved. This technique of interpolation is not applicable to Motion Reprojection, because waiting for the next frame to be rendered would add significant latency (see my long post here: Will DLSS 3.0 be supported in VR msfs? - General Discussion & Interests / Virtual Reality (VR) - Microsoft Flight Simulator Forums). If you applied interpolation for Motion Reprojection, sure you will great image quality, but with up to 3-4 frames of added tracking latency (assuming you are running at half frame rate). You will not be getting a good experience with that kind of latency in VR, due to head movement (even unintentional).

One can run this experiment today via a developer option in OpenXR Toolkit to add 3-4 frames of latency (the opposite of what "Shaking reduction" does) to a game running well... it is not usable with such latency.

This is why Motion Reprojection techniques rely on frame extrapolation (also known as forward reprojection), which does not need the next frame and will instead perform motion propagation using the most recent frame and the most recent tracking data (see my long explanation here: Motion Reprojection explained - General Discussion & Interests / Virtual Reality (VR) - Microsoft Flight Simulator Forums). This will give you only a single frame of added tracking latency, which will keep the experience usable. However motion propagation is much less forgiving to prediction error, which results in the visual artifacts.

Unless NVIDIA can provide DLSS Frame Generation as an extrapolation technique, I don't see it being usable for Motion Reprojection in VR. And my guess is if they did something of the sort, it would use the same technique of motion propagation using Optical Flow in input, which is something that vendors already do for Motion Reprojection today (Oculus does already, WMR might do soon). They will likely get similar outcome in case of prediction error, and therefore simular visual artifacts. So the benefits relative to existing Motion Reprojection are quite TBD IMO.

[some1]

2 hours ago, Stefem said:

it's pretty noticeable even with just few generated frames here and there but of course is better than a stutter in VR.

Motion reprojection is not a "few frames here and there". It effectively interleaves real and generated frames at even pace, while the main app is throttled to produce frames at half refresh rate (or 1/3rd, etc.). That's why you see the FPS counter step down from 90 to 45, then to 30 fps when reprojection activates. 

You can read more about it in the detailed explanations provided by mbucchia in his links. 

You may be thinking about an edge case when your app can produce all new frames below 11ms target (or whatever is the refresh rate of your headset), except it stutters once in a while for one frame, then you'll get only a few reprojected frames thrown into the mix. But that's not what happens most of the time. Most of the time we end up with real and generated frames interleaved at at least 1:1 ratio.

 

[Stefem]

3 hours ago, mbucchia said:

I don't think anybody is arguing that the image quality of DLSS Frame Generation is superior to Motion Reprojection. The issue is whether DLSS Frame Generation is applicable as a replacement of Motion Reprojection in VR.

DLSS Frame Generation uses interpolation between 2 already-rendered frames (also known as backward reprojection), which is how such great quality can be achieved. This technique of interpolation is not applicable to Motion Reprojection, because waiting for the next frame to be rendered would add significant latency (see my long post here: Will DLSS 3.0 be supported in VR msfs? - General Discussion & Interests / Virtual Reality (VR) - Microsoft Flight Simulator Forums). If you applied interpolation for Motion Reprojection, sure you will great image quality, but with up to 3-4 frames of added tracking latency (assuming you are running at half frame rate). You will not be getting a good experience with that kind of latency in VR, due to head movement (even unintentional).

One can run this experiment today via a developer option in OpenXR Toolkit to add 3-4 frames of latency (the opposite of what "Shaking reduction" does) to a game running well... it is not usable with such latency.

This is why Motion Reprojection techniques rely on frame extrapolation (also known as forward reprojection), which does not need the next frame and will instead perform motion propagation using the most recent frame and the most recent tracking data (see my long explanation here: Motion Reprojection explained - General Discussion & Interests / Virtual Reality (VR) - Microsoft Flight Simulator Forums). This will give you only a single frame of added tracking latency, which will keep the experience usable. However motion propagation is much less forgiving to prediction error, which results in the visual artifacts.

Unless NVIDIA can provide DLSS Frame Generation as an extrapolation technique, I don't see it being usable for Motion Reprojection in VR. And my guess is if they did something of the sort, it would use the same technique of motion propagation using Optical Flow in input, which is something that vendors already do for Motion Reprojection today (Oculus does already, WMR might do soon). They will likely get similar outcome in case of prediction error, and therefore simular visual artifacts. So the benefits relative to existing Motion Reprojection are quite TBD IMO.

Yea, absolutely, if you read some of my other above reply I have said that in it's current form DLSS frame generation is not designed for VR unless NVIDIA introduce a dedicated VR mode, I was rather simply pointing out that VR reprojection is great to compensate for frame that miss the frametime target because a crap frame is better than a stutter but it isn't desirable using it to double the framerate and used DLSS frame generation as an example of something that is designed for that and produce impressively great results

3 hours ago, some1 said:

Motion reprojection is not a "few frames here and there". It effectively interleaves real and generated frames at even pace, while the main app is throttled to produce frames at half refresh rate (or 1/3rd, etc.). That's why you see the FPS counter step down from 90 to 45, then to 30 fps when reprojection activates. 

You can read more about it in the detailed explanations provided by mbucchia in his links. 

You may be thinking about an edge case when your app can produce all new frames below 11ms target (or whatever is the refresh rate of your headset), except it stutters once in a while for one frame, then you'll get only a few reprojected frames thrown into the mix. But that's not what happens most of the time. Most of the time we end up with real and generated frames interleaved at at least 1:1 ratio.

 

I'm arguing that due to the considerable quality degradation is not a desirable solution for doubling the fps, not that it isn't capable of that and what you just said make no sense, compensating for frame that miss the target isn't an edge case and doubling the framerate isn't the norm for the simple fact that it all depend on the machine that run the software and its graphical settings.

[einarabelc5]

On 1/6/2023 at 1:45 PM, some1 said:

People may confuse two things. The Frame Generation feature introduced in DLSS3 for RTX4000 series cards will not work in VR. But the DLSS itself (antialiasing and upscaling) works just fine. It may not produce image quality to your liking, but that's another matter. 

I don't think AI based frame generation would be good for some scenarios like BFM but for Ground where the GPU usually strains and there are no sudden changes is welcome.

[westr]

So forgive my ignorance on this subject I have a Nvidia 3080 just seen in that 2.9 will likely gain DLSS support. So will this be an option you select in options graphics settings? Will it improve Anti-aliasing in VR? Will it improve VR performance? Any knowledge would be appreciated. 

[VirusAM]

So forgive my ignorance on this subject I have a Nvidia 3080 just seen in that 2.9 will likely gain DLSS support. So will this be an option you select in options graphics settings? Will it improve Anti-aliasing in VR? Will it improve VR performance? Any knowledge would be appreciated. 

Yes, as with any other game supporting dlss, it will be a graphic option.
The aim of these technologies is to improve performance...how much?it depends and it will be different per user

[Nedum]

If you are looking for more frames, because you want to raise up your picture quality, DLSS is the very wrong answer.

DLSS will >>>always<<< add smear to the whole picture. And DLSS will add artifacts and ghosting. Some more visible, some less.

Using DLSS in VR, it's like a firefight with gas. You think you will get more FPS to raise the picture quality? NO, you will get more FPS because the picture quality will be downgraded and if you want to get the good old picture quality back, you have to raise the game settings in a way the FPS will be less as before without using DLSS.

DLSS is a big No-No for VR games. Look how "good" DLSS works with MSFS. If I set the picture quality lower inside the game, the FPS are the same as with DLSS, and at the same time the picture quality is 100 times better as with DLSS.

I can't wait for all the threads about "How bad DCS looks like, even so the FPS are perfect." And the whole support team have to ask: "Do you use DLSS?" Answer: "Oh, yes I do. But I've never thought DLSS looks so bad. I can't even read my displays." Really, can't wait for this. Funny days incoming.

Edited September 30, 2023 by Nedum

[nikoel]

1 hour ago, Nedum said:

If you are looking for more frames, because you want to raise up your picture quality, DLSS is the very wrong answer.

DLSS will >>>always<<< add smear to the whole picture. And DLSS will add artifacts and ghosting. Some more visible, some less.

Using DLSS in VR, it's like a firefight with gas. You think you will get more FPS to raise the picture quality? NO, you will get more FPS because the picture quality will be downgraded and if you want to get the good old picture quality back, you have to raise the game settings in a way the FPS will be less as before without using DLSS.

DLSS is a big No-No for VR games. Look how "good" DLSS works with MSFS. If I set the picture quality lower inside the game, the FPS are the same as with DLSS, and at the same time the picture quality is 100 times better as with DLSS.

I can't wait for all the threads about "How bad DCS looks like, even so the FPS are perfect." And the whole support team have to ask: "Do you use DLSS?" Answer: "Oh, yes I do. But I've never thought DLSS looks so bad. I can't even read my displays." Really, can't wait for this. Funny days incoming.

 

There are people are using FSR 1.0 with VR. Some people supersample from 100% resolution using the the up-scaler for a better [to them] image

DCS suffers from shimmering even with MSAA; DLSS may achieve better image quality because it uses different technology to reduce this shimmer. Activating it to increase the resolution from 100% to say 140% with similar performance and reduced shimmer would be a win to many

You may be right, however without having seen what the implementation that ED has gone for, you have no idea what the image quality will actually be like

[Mr_sukebe]

11 hours ago, Nedum said:

If you are looking for more frames, because you want to raise up your picture quality, DLSS is the very wrong answer.

DLSS will >>>always<<< add smear to the whole picture. And DLSS will add artifacts and ghosting. Some more visible, some less.

Using DLSS in VR, it's like a firefight with gas. You think you will get more FPS to raise the picture quality? NO, you will get more FPS because the picture quality will be downgraded and if you want to get the good old picture quality back, you have to raise the game settings in a way the FPS will be less as before without using DLSS.

DLSS is a big No-No for VR games. Look how "good" DLSS works with MSFS. If I set the picture quality lower inside the game, the FPS are the same as with DLSS, and at the same time the picture quality is 100 times better as with DLSS.

I can't wait for all the threads about "How bad DCS looks like, even so the FPS are perfect." And the whole support team have to ask: "Do you use DLSS?" Answer: "Oh, yes I do. But I've never thought DLSS looks so bad. I can't even read my displays." Really, can't wait for this. Funny days incoming.

 

Have you tried DLAA?  If not, give it a try in that Civ sim.