Driver level DLSS

[Stahlmann]

So i've been thinking: DLSS has been independent of "game training" since 2.0 and they even integrated it into Unreal Engine so indie studios can also make use of it without working with Nvidia directly.

 

What is keeping Nvidia from just running DLSS at a driver level and forgo any specific game integration whatsoever?

 

Like with G-Sync they could just include a global toggle or add it as a 3D settings option so you could decide if a game shouldn't use it.

 

That would also clear up the mess that they introduced with every DLSS supported game coming with a slightly newer/different revision. Why not just keep it packaged and updated with the rest of the driver and leave it at that?

[jaslion]

I think it might have mostly to do with marketing?

 

Now you label your game as "Nvidia DLSS compatible". Compared to amd who has Fidelity fx and it works on anything but isn't labelled anywhere in the game.

 

[Stahlmann]

16 minutes ago, jaslion said:

I think it might have mostly to do with marketing?

 

Now you label your game as "Nvidia DLSS compatible". Compared to amd who has Fidelity fx and it works on anything but isn't labelled anywhere in the game.

FSR doesn't work universally. It also has to be worked into the game.

 

Nvidia could save significant amounts of money and time with making DLSS an universal toggle. Same with AMD and FSR for that matter.

[jaslion]

8 minutes ago, Stahlmann said:

FSR doesn't work universally. It also has to be worked into the game.

 

Nvidia could save significant amounts of money and time with making DLSS an universal toggle. Same with AMD and FSR for that matter.

Fsr works in any game you just have to unlock the functionality with a tool. I did get that mixed up but in things amd stated they said it works for virtually any game and it does since it's basically a post processing layer.

 

I do see another reason why dlss is a bit limited and that is training of the AI behind it. Probably has a thing with unreal engine about sending back data for training in different styles. Since this is actual ai upscaling and not just a post processing layer.

 

Then again their shield tv thing does it for anything so who knows.

[YoungBlade]

DLSS does require game makers (or at least engine makers) to implement because it relies on motion vectors from objects in the game. That data is not visible to an external program and may not be that easy to gather depending on the engine used.

 

FSR, on the other hand, just needs to be put between the image processing stage and when the HUD is overlayed so the text and HUD stuff doesn't look like crap. That tech should actually be trivial to add to games, so hopefully more studios see a benefit of doing. I've already seen game mods for FSR to give you an idea of how little work a full-blown studio would need to do.

[porina]

1 hour ago, Stahlmann said:

What is keeping Nvidia from just running DLSS at a driver level and forgo any specific game integration whatsoever?

One reason is that you don't scale everything. Well, you could, but games do better than that. Even before DLSS when games had (and many still have) their own render scale tech, it is usually only the world content that is scaled, but not other features like UI which remain at native. So that helps with the impression of the overall output being at native resolution.