News AMD Drops FSR 2.0 Source Code, Takes Shots at DLSS and XeSS

[renz496]

Than FSR 1.0 perhaps, not 2.0. I stopped caring about 1.0 as soon as 2.0 launched.

it is still an issue even with FSR 2.0.

 

[-Fran-]

it is still an issue even with FSR 2.0.

uh?

View: https://youtu.be/s25cnyTMHHM?t=864


Regards.

 

[digitalgriffin]

with AMD solution there is no AI being calculated even on shader cores. why DLSS end looking much better at lower resolution than FSR? because the AI part augment the upscaling on the missing data part.

DLSS is a bit of a secret sauce. But it uses key image elements from the game and compares them to known images from a super upscaled game render. So it's using image tensor cores for image recognition. ie: Is this a backpack (image recognition)? What is it's z-depth (where typically found)? What's the motion vector like? (does it move like we expect from the image training) The last is common to mpeg motion vector encoding. Then the tensor core goes "Okay we have a match" And then it says "This is what the upscaled backpack looks like. Lets apply it to the image"

This is also partially why there are things like shimmering on downscale. Because a high res scale down with questionable vector being a rough guess, it will cause shimmering. A key example of this was the encoding used on an mpg movie where there are bars or check patterns. The shimmer is awful. But it's because the image processor isn't applying temporal AA. There is more than one type of algorithm to eliminate temporal noise. (I have my favorites) So again, this is a bit of a secret sauce. I imagine DLSS2.0 is also applying a temporal element to prevent this shimmer. If the high quality source has a temporal difference of X for object Y, then the resampled version shouldn't have a temporal difference of X +- Z% for object Y.

 

[Madkog]

But if you speak philosophical, true AI won't exist until a computer has the ability to add to it's skillset dynamically on it's own. (ie: Computer: "I don't know painting. Let's learn painting today by looking at paintings") It has to do so of it's own impetus.

From what I understand on how we learn is that we learn by copying others and taking in experiences, information from others via media and trial and error. All based on previous foundations.

So for can't really just look at a painting, and learn to paint at all like those who painted it.

Thats why two people can learn the same subject, and depending on the training turn out either competent or not. Its less about intelligence and more about the training.

Tangent

This is why talent does not exist. Talented help optimise and make others talented, and time and dedication manifest the results.

 

[KananX]

This is why talent does not exist. Talented help optimise and make others talented, and time and dedication manifest the results.

No, talent is a affinity for a person to learn something with higher effectivity often unreachable by others who don’t share this affinity. Some people are so talented in something, others can try for thousand years and will never reach the same level. It’s funny to me you think talent does not exist. I can bring way more reasons and examples why it does in fact, this is just a short.

You’re just talking about education, nothing else.

 

[renz496]

uh?

View: https://youtu.be/s25cnyTMHHM?t=864


Regards.

Tim himself said at 1080p DLSS is better in that video

 

[-Fran-]

Tim himself said at 1080p DLSS is better in that video

Ah, I knew you'd nitpick. That's why I left it at that time.

Have it your way then.

Regards.

 

[KananX]

Tim himself said at 1080p DLSS is better in that video

DLSS isn’t perfect at 1080p either, I wouldn’t even use either at 1080p, simply skip ray tracing it’s not hard having a GPU that can display 1080p without RT on in native. If you don’t have the money to have a strong enough gpu for 1440p, you’re essentially a mid budget gamer, I don’t see why you need ray tracing gimmicks then, and otherwise a lot of GPUs are fast enough for 1080p.