DLSS, DLAA, thanks but going back to MSAA

[mac22]

8 hours ago, Sile said:

VR tab of the settings. „Pixel Density“

Oh and I learned its only DLSS then, since DLAA gets greyed out once DLSS get‘s selected.

 

 

found it and set it to 1.5 (looks very good) - so do you mean U have to run DLAA when DLSS is selected?

[Sile]

No. Only DLSS is running. Cannot run both. If you want only DLAA, you need to disable DLSS.

[mac22]

5 hours ago, Qcumber said:

Try pushing PD up to 1.5 and try QVFR if you are not already using it. It can produce some very good results depending on your CPU/GPU combination. 

so am running a 7950x3d and 4090(with the crystal)

selected PD of 1.5 and running DLSS and DLAA with SSS disabled with Quad - visually the results are amazing have never seen such clear cockpits - only issue is with my FPS which stays  around 45/50 - was expecting the DLSS with Quad might give me a FPS closer to 90? (note pre 2.9 quad was giving me an extra 20 fps!)

 

2 minutes ago, Sile said:

No. Only DLSS is running. Cannot run both. If you want only DLAA, you need to disable DLSS.

 

Ok so if I run DLSS do I need to disable DLAA or can leave it untouched and it will simply not run(greyed)......

see my post above for my results  - visually the result was amazing but I don't seem to be getting much benefit from DLSS or quad?

[Hiob]

11 minutes ago, mac22 said:

so am running a 7950x3d and 4090(with the crystal)

selected PD of 1.5 and running DLSS and DLAA with SSS disabled with Quad - visually the results are amazing have never seen such clear cockpits - only issue is with my FPS which stays  around 45/50 - was expecting the DLSS with Quad might give me a FPS closer to 90? (note pre 2.9 quad was giving me an extra 20 fps!)

 

 

Ok so if I run DLSS do I need to disable DLAA or can leave it untouched and it will simply not run(greyed)......

see my post above for my results  - visually the result was amazing but I don't seem to be getting much benefit from DLSS or quad?

Just leave it untouched. But honestly - with your setup, I would run DLAA instead of DLSS. If necessary tune down the PD.

[mac22]

56 minutes ago, Hiob said:

Just leave it untouched. But honestly - with your setup, I would run DLAA instead of DLSS. If necessary tune down the PD.

actually the pixel density of 1.5 was just a recommendation I got to get rid of the 2.9 blurrys and it certainly works (cockpits are amazing!) but not the fps (guessing this is what might be impacting it......)

assuming if I select DLSS all it is doing is potentially giving me extra fps? (but by selecting it assume I will get no DLAA or AA?)

 

[Hiob]

Just now, mac22 said:

actually the pixel density of 1.5 was just a recommendation I got to get rid of the 2.9 blurrys and it certainly works (cockpits are amazing!) but not the fps (guessing this is what might be impacting it......)

assuming if I select DLSS all it is doing is potentially giving me extra fps? (but by selecting it assume I will get no DLAA or AA?)

 

DLSS - lesser picture quality for the sake of more fps.

DLAA - better picture quality for the price of a few fps (roughly comparable to MSAA 2x in performance drop, but better picture than MSAA 4x IF(!) your overall fps are decent).

[Hiob]

3 minutes ago, mac22 said:

(but by selecting it assume I will get no DLAA or AA?)

DLSS is rendering the scene in a lower resolution than what is displayed and uses sophisticated upscaling techniques to enhance the picture. That includes some amount of AA. AA and upscaling (as provided by the Nvidia DL technology) are very similar algorithms.

[Werewolf_fs]

DLAA is working by default wint DLSS, if works togerther ALWAYS if DLSS is active and, I have the same quality loss with DLSS and with DLAA without DLSS, the one that creates that loss is the AA DLAA filter in both cases.

[Eugel]

Yeah, it doesn´t really make much sense to set PD to 1.5 and then run DLSS.
Might as well leave PD at 1.0 and just run DLAA.

[Hiob]

55 minutes ago, Werewolf_fs said:

DLAA is working by default wint DLSS, if works togerther ALWAYS if DLSS is active and, I have the same quality loss with DLSS and with DLAA without DLSS, the one that creates that loss is the AA DLAA filter in both cases.

DLSS and DLAA do the exact opposite. So no, they work either or and not together. One reduces the internal render resolution and upscales the drawn image, the other uses the upscaling alogorithms to render upscaled edges and lines from the native resolution.

Please do me a favour and consult the nvidia homepage or wikipedia if you don't believe me.

Imagine you have a slider for performance vs. visual fidelity. DLSS draws the slider to the left and DLAA draws THE SAME slider to the right. So how on earth can the work together? They just use the same algorithms and hardware to achieve opposite results.

https://en.wikipedia.org/wiki/Deep_learning_anti-aliasing

Edited October 30, 2023 by Hiob

[Nedum]

On 10/24/2023 at 8:58 PM, AdrianL said:

DCS is using  DLSS 3.5 but without Frame Generation 

See 

 

How so? DLSS 3.5 != DLSS -3.xx

DLSS is a word game by nVidia. It has nothing new until you use Raytracing. DLSS 3.5 without Raytraycing is DLSS 2-3. Nothing more. So without Raytraycing, and DCS has none, how is DCS using DLSS 3.5?

DLSS 3.5 isn't DLSS. Period!

[Hartsblade]

Just throwing in my input here as well.  After a bunch of testing and multiple setting changes and iterations of settings (with restarting the sim in between each time). I've decide to stick with MSAA.  I was seeing a mild performance increase using DLAA, but the image quality was much worse. Overall less clear , even with Sharpening, and a good deal of stuttering, especially down low over terrain.

[Sile]

vor 1 Stunde schrieb Eugel:

Yeah, it doesn´t really make much sense to set PD to 1.5 and then run DLSS.
Might as well leave PD at 1.0 and just run DLAA.

DLAA and DLSS introduce blurriness compared to MSAA that can be countered by Upscaling (PD up).
And DLSS gives more performance headroom for upscaling than DLAA. With PD 1.5 DLSS looks much clearer than PD1.0 DLAA in my case with lower frametime. (plus sharpening)

After lots of optimizing and then comparing to my 2.8 settings i admit MSAA still has the clearer picture. 
But i still prefer DLSS and PD 1.5 because of the shimmering of MSAA. Performance (frametime) is also 1ms better with DLSS and PD 1.5 compared to MSAA 4x in my case.
 

Edited October 30, 2023 by Sile

[Qcumber]

50 minutes ago, Hiob said:

DLSS and DLAA do the exact opposite. So no, they work either or and not together. One reduces the internal render resolution and upscales the drawn image, the other uses the upscaling alogorithms to render upscaled edges and lines from the native resolution.

Please do me a favour and consult the nvidia homepage or wikipedia if you don't believe me.

Imagine you have a slider for performance vs. visual fidelity. DLSS draws the slider to the left and DLAA draws THE SAME slider to the right. So how on earth can the work together? They just use the same algorithms and hardware to achieve opposite results.

https://en.wikipedia.org/wiki/Deep_learning_anti-aliasing

 

This statement from NVIDIA suggests that DLSS uses DLAA for anti-aliasing. 

Deep Learning Anti-aliasing

Provides higher image quality for all GeForce RTX GPUs with an AI-based anti-aliasing technique. DLAA uses the same Super Resolution technology developed for DLSS, reconstructing a native resolution image to maximize image quality.

1 minute ago, Sile said:

DLAA and DLSS introduce blurriness compared to MSAA that can be countered by Upscaling (PD up).
And DLSS gives more performance headroom for upscaling than DLAA. With PD 1.5 DLSS looks much clearer than PD1.0 DLAA in my case with lower frametime. (plus sharpening)

After lots of optimizing and then comparing to my 2.8 settings i admit MSAA still has the clearer picture. 
But i still prefer DLSS and PD 1.5 because of the shimmering of MSAA. Performance (frametime) is also 1ms better with DLSS and PD 1.5 compared to MSAA 4x in my case.
 

 

Yes. I find the same. DLSS with 1.5 PD looks better than DLAA with PD 1. With similar performance. Is because the artifacts are less pronounced with a higher res? 

[Hiob]

4 minutes ago, Qcumber said:

This statement from NVIDIA suggests that DLSS uses DLAA for anti-aliasing. 

Deep Learning Anti-aliasing

Provides higher image quality for all GeForce RTX GPUs with an AI-based anti-aliasing technique. DLAA uses the same Super Resolution technology developed for DLSS, reconstructing a native resolution image to maximize image quality.

Read carefully. The statement says that DLAA uses the same technology as DLSS. Which is exactly what I said earlier.

[Werewolf_fs]

hace 1 hora, Hiob dijo:

DLSS y DLAA hacen exactamente lo contrario. Entonces no, funcionan juntos o no. Uno reduce la resolución de renderizado interna y mejora la imagen dibujada, el otro utiliza los algoritmos de mejora para renderizar bordes y líneas mejorados desde la resolución nativa.

Por favor, hazme un favor y consulta la página de inicio de nvidia o wikipedia si no me crees.

Imagine que tiene un control deslizante para el rendimiento frente a la fidelidad visual. DLSS dibuja el control deslizante hacia la izquierda y DLAA dibuja EL MISMO control deslizante hacia la derecha. Entonces, ¿cómo pueden trabajar juntos? Simplemente utilizan los mismos algoritmos y hardware para lograr resultados opuestos.

https://en.wikipedia.org/wiki/Deep_learning_anti-aliasing

 

DLSS 3 and 3.5, use a DLAA technique as default antialiasing method. Nvidia itself say that 

Edited October 30, 2023 by Werewolf_fs

[Qcumber]

1 hour ago, Hiob said:

Read carefully. The statement says that DLAA uses the same technology as DLSS. Which is exactly what I said earlier.

Cheers. I'm beginning to get this now I think. Please correct me if I am wrong. DLAA essentially uses the same anti-aliasing technique as DLSS but without any supersampling. Have I got this right? 

[Hiob]

50 minutes ago, Qcumber said:

Cheers. I'm beginning to get this now I think. Please correct me if I am wrong. DLAA essentially uses the same anti-aliasing technique as DLSS but without any supersampling. Have I got this right? 

I think you need to understand that Anti-Aliasing (for our purposes that means getting rid of staircases) is based on super sampling. Very very basically it means that a new (previously non existent) data point is interpolated from two existing points. The existing points can be different from each other by resolution (super-), by time (temporal-) or different rendering-passes (multi-). Obviously it’s more complicated than this, but that should give an idea.

DLSS renders in a lower resolution to save performance and then upscales back to the output resolution. It applies its deep learning super-sampling algorithms to save the quality during this process. 

DLAA which was introduced later based on the same algorithms does basically the same but renders from the native (or even higher) resolution and improves the quality over native resolution.

It is two opposite directions on the same spectrum if that makes any sense.

I‘m really struggling here to put the idea into the right words. English is not my native language. 

If you really want to know the details, you‘d probably best find and read the white-papers from Nvidia on the technology.

[skywalker22]

16 minutes ago, Hiob said:

I‘m really struggling here to put the idea into the right words. English is not my native language. 

I would dare to say that not even a native english speakers cannot put this correctly, since its so complicated 

But you explained it will, at least I got the point. thx

[Qcumber]

1 hour ago, Hiob said:

If you really want to know the details, you‘d probably best find and read the white-papers from Nvidia on the technology.

Cheers. I'll take a look. I'm not a computer expert but have a background as a scientist so hopefully some of it will make sense.