DLSS & VR

[av8orDave]

With today's newsletter indicating the DLSS isn't too far off in the distance, does anyone have any insight into what (If anything) DLSS will mean for VR in DCS? Just curious.

 

[Marshallman]

Following~!!

 

[Tony86]

same

[Qcumber]

[Timmy51]

I think it should work fine in VR as it works in MSFS

[Willie Nelson]

47 minutes ago, Timmy51 said:

I think it should work fine in VR as it works in MSFS

I don’t know of anybody having a particularly good experience in VR with DLSS in that other non military sim. I’ve tried it too. Barely noticeable FPS improvement if any and clarity sacrifices all around. I hope ED have done something better than the other mob. 

[diego999]

3 hours ago, Willie Nelson said:

I don’t know of anybody having a particularly good experience in VR with DLSS in that other non military sim. I’ve tried it too. Barely noticeable FPS improvement if any and clarity sacrifices all around. I hope ED have done something better than the other mob. 

 

This. Tried on FS and noticed a lot of artifacts while the performance gains were minimal.

Let's wait and see but as far as I've seen, it might help a bit but it won't be the holy grail some people are expecting.

[Mr_sukebe]

Please try DLAA

[mbucchia]

4 hours ago, Mr_sukebe said:

Please try DLAA

Isn't DLAA just DLSS without upscaling? Ie: you won't gain any perfomance, you will actually lose a little bit, because the game won't render any faster (same resolution) and you are actually adding an extra AA pass.

So it all depends what you are trying to achieve, but AFAICT, if you are seeking performance, DLAA isn't for you.

[Mr_sukebe]

Absolutely, DLAA is just an alternative to MSAA, and deliberately avoids upscaling.

Having said that, I found it preferable to MSAA visually.

[Tom Kazansky]

3 minutes ago, Mr_sukebe said:

Having said that, I found it preferable to MSAA visually.

Can you tell us about the performance difference between 4x MSAA and DLAA?

[Mr_sukebe]

I didn’t conduct a performance test, it was purely that I much preferred the visual improvement of using DLAA, so much so that it was worth whatever performance degradation

[Willie Nelson]

2 hours ago, Mr_sukebe said:

Absolutely, DLAA is just an alternative to MSAA, and deliberately avoids upscaling.

Having said that, I found it preferable to MSAA visually.

But it's performance I need......! 

"I canny give it any more Jimmy, she's gonna blow!"

[freehand]

1 hour ago, Willie Nelson said:

But it's performance I need......! 

"I canny give it any more Jimmy, she's gonna blow!"

Give me full power scotty or your fired.

[MIghtymoo]

Guess we just have to wait a little longer and test ourselves in 2.9

[Mr_sukebe]

3 minutes ago, MIghtymoo said:

Guess we just have to wait a little longer and test ourselves in 2.9

Quite.

Basing any decisions and views on testing in a different game engine is really not very clever 

 

 

[Ahogephilia]

 

00:00 ~ 00:10

In my personal experience,

DLSS in MSFS causes severe artifacts, like arm of carrier crew in this video.

Airport lights at night can sometimes resemble the work of a Van Gogh or Georges Seurat.

But, FPS improvement is amazing. Even though I still prefer TAA.

 

[Nedum]

On 9/29/2023 at 6:50 PM, davidrbarnette said:

With today's newsletter indicating the DLSS isn't too far off in the distance, does anyone have any insight into what (If anything) DLSS will mean for VR in DCS? Just curious.

 

DLSS is an Upscaler. DLSS is AI driven. It will do things the engine would never do. At first, DLSS will always add smearing. If you can use DLAA, the smearing will be less visible on a normal 2D screen. You can see more ghosting and shimmer, but that's not a must.

The problem with VR is, all the issue will be seen like under a magnifying glass. You will see the issues better. In "the other sim" you can clearly see how bad VR looks with DLSS enabled.

You can't even read the displays like before. DLSS may be something for a monitor user, but not for a VR gamer, not at all.

7 hours ago, Mr_sukebe said:

Quite.

Basing any decisions and views on testing in a different game engine is really not very clever 

The issues DLSS is generating are AI based and not engine driven. Yes, you will get slightly different expressions, but the difficulties you will get, if you get them, are the very same.

There is no DLSS-game out there without all the known DLSS issues. How hard they will have an impact, that might be a question. But there is no question you will get them.

If you really believe this time, all will become different (better), I would bet my money on the opposite.

 

Edited October 1, 2023 by Nedum

[Hoirtel]

I wonder if the issues with DLSS and VR are caused when reprojection is also working. It would make sense that these two would not be compatible. Give the additional hardware requirements for DLSS it should do a better job than reproduction technology? However one of the main reasons I still use meta is because their reprojection is just so good. 90% of the time I can't tell if its on or not in DCS.

[Qiou87]

52 minutes ago, Hoirtel said:

I wonder if the issues with DLSS and VR are caused when reprojection is also working. It would make sense that these two would not be compatible. Give the additional hardware requirements for DLSS it should do a better job than reproduction technology? However one of the main reasons I still use meta is because their reprojection is just so good. 90% of the time I can't tell if its on or not in DCS.

Reprojection is similar, in concept, to DLSS frame generation (a.k.a "DLSS 3.0"). What is being discussed here, since people refer to upscaling, is DLSS 2.5 or "Super sampling", which is rendering the game in a lower resolution (to get extra FPS) then applying an upscaling using deep learning to make it look better than normal upscaling. I don't see why reprojection and upscaling wouldn't work together, however I do agree that frame generation and reprojection would potentially cause issues if used simultaneously. But here we would run in other issues, such as the fact that frame generation as implemented today adds latency (you get 100fps but the latency is actually slightly worse than if running at 50fps). Latency in VR is vomit-inducing and overall a big no-no.

I am deeply curious how DLSS will be implemented in DCS, and what we get out of it in VR. Will it be better AA through DLAA (let's face it, today's MSAA is difficult to apply for VR as it eats performance too much, and running without adds a ton of shimering), is frame generation even going to be implemented? Will DLSS in VR be less bad than in "the other sim" (my own experience reflects that of others before in this topic : it isn't doing much for framerate and visuals were horrible)? We will see.

The issue with DLSS is that it is "cleaning up" the image using deep learning, but probably doing so using standard gameplay on pancake. Once you start to have the player's viewpoint moving on 6 axis at the same time, with unpredictable head movement on all axis simultaneously, it becomes very difficult to predict the next frame and therefore use all those fancy IA techniques...

Edited October 2, 2023 by Qiou87

[some1]

12 minutes ago, Qiou87 said:

is frame generation even going to be implemented?

Nvidia frame generation requires DX12, given that DCS is still DX11 and is being converted to Vulkan, I doubt we'll see that in DCS any time soon. 

[Qcumber]

In theory DLSS should produce results better than using FSR or NIS through open xr toolkit. I find that OXRTK FSR works pretty well as long as you don't apply too much (max of about 90% of native resolution). It does create a small amount of blurring, but this can be mitigated by a bit more sharpening. Another use is to supersample above the headset native resolution to reduce aliasing. For example 4000x4000 pixels at about 80% (3120x3120) can negate the need for MSAA. I am hoping that DLSS will do a better job than FSR. Is this a realistic expectation?

[average_pilot]

My understanding is that DLSS tries to guess how the frame would look like if it were of a higher resolution based on its training data. So, depending on how representative is the training data of the frame it is processing at any moment it introduces more or less visual artifacts on the final result. That's basically the problem of this approach compared to traditional deterministic image processing. Again, just my understanding, I'm happy to be corrected if something is wrong.

(Edit: typo)

Edited October 5, 2023 by average_pilot

[Hoirtel]

6 hours ago, Qiou87 said:

Reprojection is similar, in concept, to DLSS frame generation (a.k.a "DLSS 3.0"). What is being discussed here, since people refer to upscaling, is DLSS 2.5 or "Super sampling", which is rendering the game in a lower resolution (to get extra FPS) then applying an upscaling using deep learning to make it look better than normal upscaling. I don't see why reprojection and upscaling wouldn't work together, however I do agree that frame generation and reprojection would potentially cause issues if used simultaneously. But here we would run in other issues, such as the fact that frame generation as implemented today adds latency (you get 100fps but the latency is actually slightly worse than if running at 50fps). Latency in VR is vomit-inducing and overall a big no-no.

I am deeply curious how DLSS will be implemented in DCS, and what we get out of it in VR. Will it be better AA through DLAA (let's face it, today's MSAA is difficult to apply for VR as it eats performance too much, and running without adds a ton of shimering), is frame generation even going to be implemented? Will DLSS in VR be less bad than in "the other sim" (my own experience reflects that of others before in this topic : it isn't doing much for framerate and visuals were horrible)? We will see.

The issue with DLSS is that it is "cleaning up" the image using deep learning, but probably doing so using standard gameplay on pancake. Once you start to have the player's viewpoint moving on 6 axis at the same time, with unpredictable head movement on all axis simultaneously, it becomes very difficult to predict the next frame and therefore use all those fancy IA techniques...

 

I know we are not getting 3.0 frame gen DLSS, I am wondering about the upscaling. If the GPU is upscaling frames that are already "guesstimated" by the VR software that might be a recipe for problems. Considering reprojection already causes artifacts AI upscaling would probably make these a bit worse.

I think the future of VR is probably a version of DLSS and move away from reprojection, but maybe not with current versions of DLSS. 

I do have a good bit of faith that ED is working hard to support VR these days so I expect what is possible they will do. Interesting to see what that is. 

Edited October 2, 2023 by Hoirtel

[St4rgun]

Here's my 2cents based on experience with MSFS and Reverb G2:

DLSS is the BEST visual quality in that sim in VR, period. The trick is that in VR you'll need quite a bit of supersampling for not only anti-aliasing but for correctly counter-distort the lenses. That's why the G2's 100% resolution set to around 3100x3100 pixels per eye in Steam, despite the headset having "only" 2160x2160 pixels per eye resolution.

The best we can do is set the headset to 3500x3500 or even 4000x4000 resolution per eye (sounds silly, eh?), and turn on DLSS to have the game's rendering lower (like the native 2160x2160) and let DLSS do the proper upscaling. The result will be not only a MUCH sharper image, but a much larger hotspot (even edge to edge clarity). It's incredible.

I have very high hopes in DLSS an DLAA in DCS with VR, I'm super excited!