NVIDIA Bringing DLSS 2.0 Support to UE 5, Linux via Proton, and Even More Titles

[medi01]

because DLSS 2.0 uses tensor cores

Because tensor cores is anything but a bunch of FPs.
And because we do not have cards that could do more FPs than 2060 (including its tensor ops)

Come on, people.

FSR drivers

I'm not getting it...
It shouldn't be a driver level thing, although, I'd understand that drivers could speed up certain aspects of it.

 

[wolf]

WHY DON'T YOU BRING DLSS DOWN TO US PEASANTS RUNNING GTX 16 SERIES CARDS!

Why hasn't this been done yet? Seriously. Is it a hardware issue, or just Nvidia being assholes?

non-RTX cards lack the hardware that DLSS runs on, but don't worry you'll get FSR.

 

[zlobby]

My end goal is to not run Windows at all.

Which is admirable. No sarcasm here. But in case you need it for some API or a game (your words from another post) you can always run it well-isolated from other critical stuff.

 

[Soul_]

"texture upscaling" huh? That is how we refer to TAA derivatives these days?

LOL!!! Easy there Joker, finding the right pixels to upscale with AI is far closer to real image than TAA could ever predict, we both know it. NVidia had those pixel prediction and photo upscaling working on Pascal long time before DLSS. It is just that now they can apply it realtime in the frame generation pipeline.

Every DLSS 2.0 title I have played on Quality mode is just like native. I can't tell native vs DLSS 2.0 quality.

Like they say, proof is in the pudding.

 

[medi01]

finding the right pixels to upscale with AI

pixel prediction

photo upscaling

apply it realtime

Amazing stuff, isn't it?

 

[Soul_]

Because tensor cores is anything but a bunch of FPs.

Yes they are FPs, but running lower precision on existing FPs with double or quad the clock-rate is not something that would happen magically. Tensor cores are FPs that can run at big multiples of clockrates, unlike the traditional FPs. So yes, they are special and they are the reason why lower precision calculation are so fast on these chips and are essentially required for anything realtime-RT or AI related. They are one of the reason why AMD is not upto the scratch on RT performance, even when they are fast in rasterization.

And because we do not have cards that could do more FPs than 2060 (including its tensor ops)

Come on, people.

It is not about having FPs, it is about having FPs which can run much faster than traditional FPs due to clockrate multiple. Cmon Man!!!

 

[medi01]

It is not about having FPs, it is about having FPs which can run much faster than traditional FPs

Oh, boy, am I still on a tech forum...

 

[medi01]

Reviews are out:

 

[R-T-B]

Which is admirable. No sarcasm here. But in case you need it for some API or a game (your words from another post) you can always run it well-isolated from other critical stuff.

The API/game front is slowly being conquered. Yes there is a performance penalty but at this point I could accept it.

It's the HDR support that prevents me from my end goal. That's it.

 

[95Viper]

Please remember the Guidelines of the Forums.
If needed, please review them if you need to.

Posting in a thread
Be polite and Constructive, if you have nothing nice to say then don't say anything at all.
This includes trolling, continuous use of bad language (ie. cussing), flaming, baiting, retaliatory comments, system feature abuse, and insulting others.
Do not get involved in any off-topic banter or arguments. Please report them and avoid instead.

Reporting and complaining
All posts and private messages have a "report post" button on the bottom of the post, click it when you feel something is inappropriate. Do not use your report as a "wild card invitation" to go back and add to the drama and therefore become part of the problem.

 

[Batailleuse]

All of those that support 2.0, you just have to update the file yourself. Devs will probably do that when a patch comes out.

Even 2.0 is simply the best upscaling solution out there. Nothing comes close.

yeah

physX was the best (semi dead/replaced by open version)
Nvidia 3D vision was the best (its dead)
SLI was the best (its dead in the water except for computing)
and Gsync was the best (not dead per se, but Freesync reign master on the market now, Gsync was just simply too expensive and is now just a halo tech used on a few monitors that cost over 1000$)


Nvidia sure does have nice cool new things sometimes, but when its locked to specific more expensive hardware then it does not persevere.

will most likely be the same for DLSS and RTX

both those can and will be done by engines, with different names that is but same or close or better without locking you with a specific hardware. and that is why on long term Nvidia always lose the market share battle on their techs.

UE5 has lumen (rtx) and dlss like features for instance