Can someone explain DLSS to me....

This is how I understand it:
There is a "server" with a reference 16k image.
It lays that image against what you are seeing and then tries to make your (low res) image look as much as possible like the 16k image.
So does that mean you have to be online to even get DLSS support and/or to help it make the DLSS support for the game you are playing better?

Also why does DLSS On improve performance?
Does it mean that when you select 4k res in a game with DLSS On you are actually running it just on for example 1080p but DLSS "upscales" it to 4k with good quality?

And also, if this is how it works, why do games even need to support it? why cant Nvidia just make a 16k reference image themselves and have that communicate with the users to train the "server" in using it.
Why would this not be something you can just turn on in the Nvidia Control Panel for every game?

ZoneDymoCan someone explain DLSS to me....

This is how I understand it:
There is a "server" with a reference 16k image.
It lays that image against what you are seeing and then tries to make your (low res) image look as much as possible like the 16k image.
So does that mean you have to be online to even get DLSS support and/or to help it make the DLSS support for the game you are playing better?

Also why does DLSS On improve performance?
Does it mean that when you select 4k res in a game with DLSS On you are actually running it just on for example 1080p but DLSS "upscales" it to 4k with good quality?

And also, if this is how it works, why do games even need to support it? why cant Nvidia just make a 16k reference image themselves and have that communicate with the users to train the "server" in using it.
Why would this not be something you can just turn on in the Nvidia Control Panel for every game?

It's done "manually" by Nvidia using their super duper rig for each individual game to ensure everything looks good. The results are then incorporated into the drivers and replicated locally using your Nvidia GPU.

Vayra86nvidia/comments/9g65w4

ZeppMan217It's done "manually" by Nvidia using their super duper rig for each individual game to ensure everything looks good. The results are then incorporated into the drivers and replicated locally using your Nvidia GPU.

Right so what I was thinking then is more or less correct.
And its up to Nvidia then to choose which games the system learns before rolling it out to the consumer, so some games may never be chosen by them, that is kinda sad to think about.

ZoneDymoso some games may never be chosen by them, that is kinda sad to think about

Look on the bright side, that's only an issue at the moment. As tech and soft mature, it might become universal, like image sharpening or integer scaling.

ZeppMan217Look on the bright side, that's only an issue at the moment. As tech and soft mature, it might become universal, like image sharpening or integer scaling.

I was thinking of some Folding at Home kinda deal where all Nvidia users could help, but someone (Nvidia probably) would still have to get those super high quality images first for it to work.

where can we find some real comparison shots (NOT the one marketed by nGreedia) with the new DLSS 2.0 for PC?

ZoneDymoRight so what I was thinking then is more or less correct.
And its up to Nvidia then to choose which games the system learns before rolling it out to the consumer, so some games may never be chosen by them, that is kinda sad to think about.

Yes so far everything AI and deep learning translates into a titanic amount of effort for minimal gains, and certainly not 'self learning' in the pure sense of the word. Unless endless crunching on a per-case basis is somehow intelligent. :P

ZoneDymoAnd also, if this is how it works, why do games even need to support it? why cant Nvidia just make a 16k reference image themselves and have that communicate with the users to train the "server" in using it.
Why would this not be something you can just turn on in the Nvidia Control Panel for every game?

Here is the less technical explanation :

The game is run on a render farm at very high resolutions and stills from the game are fed into a program that tries to generate a model that generalizes how the images are supposed to look. This model is then sent through a driver update to the machine where it is fed the images from the game running locally at lower resolutions and hopefully it can then scale up the images to look like the original high resolution images that were generated back on the render farm.

Emphasis on the word "hopefully".

You don't have to be connected to the server, you just need to download the model once (ideally). The model differs from game to game for accuracy purposes, you could make one global model but the results are going to be worse.

Vayra86nvidia/comments/9g65w4

When DLSS 2x is used exclusively to enhance image quality it gives us an image almost the same as 64x SS. "

:roll:

Someone wake me up if they get to that point before abandoning DLSS.

Prince Valiant:roll:

Someone wake me up if they get to that point before abandoning DLSS.

Do keep in mind, you're talking about the company that pretty much controls discrete GPU progress in this market.

I agree, the per title optimization is utterly disgusting and useless. But this is Nvidia. Look at their drivers. They bring day one game ready stuff every time;; for reasons that vary, but its still there and they don't really miss a lot of titles at all. And in terms of pushing the performance envelope.... I do think this is the direction in the near future anyway if we want more faster bigger. Those nanometers won't get much smaller and bigger chips are not for everyone's wallet, and is completely counterproductive when growth is the norm. If you need more volume, you need smaller dies.