• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The RT Related Games, Benchmarks, Software, Etc Thread.

Status
Not open for further replies.
nvidiamd said:
In theory, while moving, you will get lower quality image ( a lot of blur ) even on native if it is using TAA, for the same reason you will get lower image quality when using DLSS. But for something like DLSS perf, the image degradation should be bigger because it is not doing only aliasing but also upscaling for a much lower res.
Click to expand...

Some of that will depend on how big the delta is between the previously rendered frame and the current one as (I believe) DLSS uses prior frame(s?) to construct the current one (part of the temporal component, no?)

Wikipedia said:
Nvidia's DLSS operates on similar principles to TAA. Like TAA, it uses information from past frames to produce the current frame. Unlike TAA, DLSS does not sample every pixel in every frame. Instead, it samples different pixels in different frames and uses pixels sampled in past frames to fill in the unsampled pixels in the current frame.
Click to expand...

Interesting post here: https://old.reddit.com/r/hardware/comments/fvgf7d/how_dlss_20_works_for_gamers/
 
Last edited:
Aegis said:
Some of that will depend on how big the delta is between the previously rendered frame and the current one as (I believe) DLSS uses prior frame(s?) to construct the current one (part of the temporal component, no?)

So the less dramatic the change, the more 'accurate' the reconstructed image will be.
Click to expand...
Yes that is the theory. You don't move or you move very slow -> less changes in consecutive frames -> better TAA/reconstruction. You move faster -> bigger changes and more "guessing" for TAA/DLSS.
 
nvidiamd said:
Yes that is the theory. You don't move or you move very slow -> less changes in consecutive frames -> better TAA/reconstruction. You move faster -> bigger changes and more "guessing" for TAA/DLSS.
Click to expand...

It's because the changes between frames are too drastic if you move the camera fast, basically screws over the clamping step (which is where DLSS separates itself from TAAU et al). And the more reliant the technique is on the accumulation part the worse it's going to be in such scenarios. Ironically it's good for third person, more "scenic" games like open world ones (ala WD:L et al) that you'd find more at home on consoles, but worse for faster paced games played with mouse, such as Doom Eternal etc. that you'd play on PC.

Certain approaches to anti-aliasing and upscaling, DLSS 2.X included, use data from previous frames to help drive the AA/upscaling (these approaches are typically referred to as temporal approaches/techniques/methods/solutions, since they involve time, ie they're temporal), and naturally these previous frames will be different if the camera itself or other objects are moving or changing, which, if unaccounted for, will result in something like motion blur, called ghosting.

Normally this difference due to movement can be accounted for by keeping track of how the camera/objects move (called motion vectors), and basically adjusting where you look in the previous frame based on this, to sort of reverse the movement, which can fix the ghosting caused by movement.

But, sometimes this isn't enough, whether it be due to the motion vectors being inaccurate or by the scene changing in some way that's unrelated to movement (say, a shadow moved across a surface), and so even though most of the ghosting has been fixed, there's still a bit that remains on parts of the screen that weren't properly accounted for.

Normally, this is accounted for by doing something called neighbourhood clamping, where you basically look at an area in the current frame, compare that to the previous frame, and if the difference is too big, throw out the previous frame and just use the current frame, which can catch most of the remaining ghosting.

But, sometimes this neighbourhood clamping isn't perfect, and will still let a little bit remain, which is generally what people really refer to by ghosting, the little bit that survived both the reversing of movement, and the neighbourhood clamping.
Click to expand...

https://www.reddit.com/r/pcgaming/c...ound_to_reduce_ghosting_in/h3wcawk/?context=3
 
Poneros said:
It's because the changes between frames are too drastic if you move the camera fast, basically screws over the clamping step (which is where DLSS separates itself from TAAU et al). And the more reliant the technique is on the accumulation part the worse it's going to be in such scenarios. Ironically it's good for third person, more "scenic" games like open world ones (ala WD:L et al) that you'd find more at home on consoles, but worse for faster paced games played with mouse, such as Doom Eternal etc. that you'd play on PC.



https://www.reddit.com/r/pcgaming/c...ound_to_reduce_ghosting_in/h3wcawk/?context=3
Click to expand...
Yes when we talk about ghosting that is the antialiasing part but DLSS and any other TAAU method will also suffer for the lack of data for upscaling if the changes between frames are too big. As they say, for ghosting you can discard the data from the previous frame, but you can't get more data for upscaling. And this should be more obvious when you do the upscaling from a lower res ( dlss performance ).
 
nvidiamd said:
Yes that is the theory. You don't move or you move very slow -> less changes in consecutive frames -> better TAA/reconstruction. You move faster -> bigger changes and more "guessing" for TAA/DLSS.
Click to expand...


This is more BS from you.

Thete isn't nore guessing when there is movement. The whole point of using motion vectors is so the pixels can be projected into the current view space and more information is accumulated, regardless of movement
 
Poneros said:
It's because the changes between frames are too drastic if you move the camera fast, basically screws over the clamping step (which is where DLSS separates itself from TAAU et al). And the more reliant the technique is on the accumulation part the worse it's going to be in such scenarios. Ironically it's good for third person, more "scenic" games like open world ones (ala WD:L et al) that you'd find more at home on consoles, but worse for faster paced games played with mouse, such as Doom Eternal etc. that you'd play on PC.



https://www.reddit.com/r/pcgaming/c...ound_to_reduce_ghosting_in/h3wcawk/?context=3
Click to expand...


The deep learning comed into this because the ML model learns a better model of history rejection.
 
Grim5 said:
Yes it looks nice but that's more because it's Naughty Dog, and they put way more effort into Hero assets, animations and mocap than any other developer in the industry and less to do with the graphical technique, which requires a lot of time, effort and resources to do and is performance intensive anyway.

I remember when games first started using this technique for mirrors in the late 90s/early 2000s a lot of people would disable it in the game menu because it would hit graphics card very hard - and developers would place these mirrors in very clean looking rooms so that the player isn't distracted to something that could break the immersion like picking up a brick and seeing it not reflected back.
Click to expand...

Yup, ray/path tracing should allow more beautiful games all round, but will take some time until we'll have a capable enough card (at least 1080p/60fps), at around $200 range
 
Red Dead Redemption 2 DLSS Patch releases on July 13th - https://www.dsogaming.com/news/red-dead-redemption-2-dlss-patch-releases-on-july-13th/

Rockstar has just announced that the DLSS patch for Red Dead Redemption 2 will release on July 13th. According to the team, both Red Dead Redemption 2 and Red Dead Online will support DLSS, allowing RTX owners to gain better performance.

It will be interesting to see what DLSS version Red Dead Redemption 2 will be using. My guess is that it will feature version 2.2.6.0, though I wouldn’t be surprised if it’s using an older version. After all, Doom Eternal’s DLSS patch featured version 2.1.66.0.

We also know that NVIDIA is working on a new DLSS Mode, the Ultra Quality Mode. However, and since we don’t have any additional details, we don’t know whether RDR2 will be the first game that will support it.

Naturally, we’ll be sure to benchmark this DLSS patch once it comes out, so stay tuned for more!
Click to expand...
 
if it fixes arthur's hair at lower resolutions i will be gloating all over it xd

only 4k can somewhat make arthur's hairs tolerable but we shall if dlls can reconstruct it even at 1080p
 
NVIDIA publicly released general version of DLSS SDK for custom engine

Noticed some interesting stuff in the reddit thread:

I found enabling Auto Exposure removes ghosting. This is interesting...

https://youtu.be/bY3DUJ-NYJ4

Edit:

In the programming guide doc:

If ExposureValue is missing or DLSS does not receive a correct value (which can vary based on eye adaptation in some game engines), DLSS may produce a poor reconstruction of the high-resolution frame with artifacts such as:

  1. Ghosting of moving objects.

  2. Blurriness, banding, or pixilation of the final frame or it being too dark or too bright.

  3. Aliasing especially of moving objects.

  4. Exposure lag.
Yet we don't know improper ExposureValue is the only factor that causes ghosting, many further experiment needed by actual developers.

Edit 2:

After some investigation, Auto Exposure is the only new feature added in DLSS 2.2 and properly documented in this release. I think game devs couldn't provide good exposure value to the DLSS runtime before 2.2, so NVIDIA added this feature.
Click to expand...

Also, a way to remove the sharpening! Thank goodness:

There's a developer version of nvngx_dlss.dll in the package that can be used to remove built-in DLSS sharpening in games that force enable it (RDR2). Hit alt+ctrl+f7 to toggle.

The file is located in this folder:

nvngx_dlss_sdk\Lib\Windows_x86_64\dev

To see a detailed onscreen indicator and hotkeys for stuff like debug screens, run ngx_driver_onscreenindicator.reg located in:

nvngx_dlss_sdk\Utils
Click to expand...

Will have to try this with RDR 2 as dlss sharpens the image way too much for my liking.
 
"If the developer enables sharpening, the level of sharpening should be controllable by the end-user. For information on how to display the user facing selection, please see the “NVIDIA RTX UI Developer Guidelines” (the latest version is on the GitHub repository in the “docs” directory)."

i guess none of the developers get the memo from nv

lets hope we're given a sharpener slider/toggle in future implementations
 
Status
Not open for further replies.
Back
Top Bottom