Nvidia DLSS, this thread has been visually enhanced.

Looks like the DLSS sourcecode was stolen and publicly released in the recent Nvidia attack. Not a big deal, but we may see it ported to run on shaders (read, AMD) if that performs well enough to be worthwhile.

Does DLSS not require hardware specific to RTX cards to be useful? Otherwise, FSR is just the same thing?

DLSS from Nvidia does require tensor cores, but there’s no reason it couldn’t run on shaders, it would just be less performant. Real question is how much less performant.

Unreal Engine 5’s TSR is basically the same thing as DLSS, except it runs on shaders and only works in UE5.

FSR is not the same thing, FSR doesn’t use motion or temporal data (previous frames) to reconstruct the image, so it does a markedly worse job.

Ahh, thanks for the clarification. I’m going to go out on a limb and say “less performant enough” as otherwise nvidia would have added it already.

Not at all. Nvidia wants you to buy a new GPU.

We’ll only know once the tech gets ported, it’s not like Nvidia has any incentive to have one of it’s flagship reasons for heir new GPU’s just run anywhere else.

Is the leaked code that useful if it requires training for each game on their superdupercomputers to get a profile out?

Presumably it includes the model and associated parameters.

AMD won’t touch the source code with a 100ft pole.

That being said, it did supposedly contain the LHR code, which may allow miners to understand the heuristics to bypass mining limitations more accurately.

Don’t think he meant AMD would use it themselves, just that someone would port it to work on AMD cards (if performance worthy). Heck - AMD is probably already working on a DLSS equivalent and now will be scared of the comparisons.

DLSS hasn’t worked that way since version 2.0, there are no individually trained games.

Well Intel is, with their XeSS. It works like DLSS, is accelerated by their upcoming hardware, but is meant to also work on Nvidia and AMD cards, just obviously not as well.

Yep, XeSS seems promising.

Rumors are that AMD will announce FSR 2.0 at GDC this month. It will be basically what everybody wanted FSR1 to be; DLSS that works on any GPU, using temporal and motion data to reconstruct the image.

Couple of XeSS videos… still to watch :)

First one apparently running on an Intel GPU.

“All files are digitally signed with NVIDIA’s signature”

I’m not sure that’s a flex anymore.

In a number of spots in those videos, we see a comparison between 1080p, and XeSS 4k. When they show that, are they showing 4k downsampled to a 1080p screen, or are they merely contrasting that native 1080p and XeSS 4k simply have about the same relative performance?

EDIT: Derp, I get it now. Nevermind :P

FSR2 might not be the same as DLSS but it’s cool it exists. I think I saw somewhere that the next AMD GPUs will have dedicated hardware for it too.

Look at that. XeSS at least is real :)

Looked nice too.

“Slower and not as not as good as DLSS” is disappointing though