NVIDIA Unreal Engine 4 Plugin Brings DLSS To More Games

The same can be said about AA in most games. Specially UE3 games ported from consoles. Often the choice is between an awful TAA implementation and FXAA crap.

 

Yes. Any AA can look great or bad depending on particularities of the implementation, complexity of assets etc.

What's more less fixed is what particular AA can realistically achieve, ie its perf/IQ profile. And thats where DLSS is aggressive and highly ambitious. Except it hiccups now and then.

But then again everything short of proper Super-Sampling does the same.
Except we've learned to live with all kinds of artifacts and have become immune to errors in traditional AA and rendering methods. So ppl say things like TAA is garbage: Yes, sometimes it is. But I would never go back to MSAA + Tr days. For myriads of reasons.
And neither will the gaming industry. And its not like everyone is on the mission to make our games look like garbage. These are the smartest ppl in the business. They made real-time Ray Tracing possible for christ sake.

 

crysis remastered pc dlss is not working no fps improvement nothing FPS keeps the same whatever settings y do

 

DoF? Why TF would DoF be applied to it? Either way it doesn't matter what the reason is for why it's crap, the fact is that's how DLSS performs in games now, as is, it can't handle motion.

I will continue to ignore all of it, because it's not reality. Results in games I play matter, not theoretical scenarios which may never happen.

Best case scenarios as in the close up detail lost is minimal, I listed those games because from the ones I've seen, which is nearly all DLSS games, they were the closest to native close up. They all still had clearly lost detail.

That's such a weird thing for them to do. Just give us full control over what resolution it scales from to what we want.

 

@Noisiv : What's the point in creating false narratives? It almost looks like you do not understand meaning of word "motion" when you "prove" DLSS being superior in motion by disabling AA for native rendering. I told you many times that way of getting images for comparison is:
- Get demo which does not run fixed fps, but renders at fixed engine time intervals. (Which means every 20ms of in-game time, one frame is rendered even if it takes minute to render. That technique is used to create in-game tech demo movies.)
- Get them in lossless format, compare.

Secondly, I have strange feeling, that you again used wrong testing procedure. So, why don't you describe what processing did happen for Reference/Native/DLSS images.
What was their rendering resolution and how it increased and decreased? What filtering was used, and if upsample or downsample, what algorithm been used.

When someone uses word reference (while yours reference does not look bad), one should show that method which led to its creation really leads to creation to image which is considered optimal. Each image lacks clear description on how it has been obtained.

And what's the point in calling people who disagree "shills"? It more looks like projection considering your posts are flaky due to missing information, and clear use of irrelevant scenarios as proofs.

 

it does get rid of most image flickering in motion though.at the cost of slight artifacts.again,a trade off that I'm willing to take,like IQ for performance increase.
and once again,whoever says dlss 2.0 is smeared only repeats false talking points.it's almost overwhelmingly detailed to me.not saying "better",but a lot of vague elements that are otherwise lost are being reconstructed.

whether dlss 2.0 is useful is not even a debate atm.really,amd fanbase,just stop repeating the same thing over and over.
the question is
1. at what point the pros start to outweigh the cons.
2. what should be the adoption rate for this to make it worth spending your money

my perspective - 1440p balanced/quality is where things start to get interesting and I would decide to use dlss more than not use it.The adoption rate tho - it's too slow.It isn't useless,but it isn't world changing atm,and it has the potential to.

 

It is good at reconstructing repeating line patterns. But decreases per pixel precision.

And as for AMD's fan base. Some of nVidia's owners are even more critical about it. And some are completely blinded by something and they see no downsides.
So, it is question of objectivity.

I liked some of AA DLSS 1.0 had. Like best I have ever seen. But then there were artifacts too. But it preserved more sharpness to edges than DLSS 2.0.
Maybe they should use ML-logic-tree to decide which parts of image will be processed by which algorithm. This pre-pass may even save on computational power as more complex procedures may not be fitting for certain input data.

 

I think DLSS is a really nice addition to the available AA solutions, and the only one i know of capable of preserving thin details (1 pixel/1 pixel line/hair/fur/all contours of object).
TAA still have room for improvements too, UE4 was mentionned earlier, there's currently an experimental new TAA available using console command, mostly for PC and next-gen consoles.

 

I did this (measured all kinds of IQ comparison numbers) for my own use. And then posted it for others.
You are free to use any method you wish to continue reviewing DLSS that you never even used. Including whatever chore you suggested to me up there.
As if discussing almost anything with you is not already a chore, you decided to give me more chores by creating "the right comparison method" for me to follow.

you're gone, far gone...

 

If you want to see something, you need right tool. You are using telescope to look at microorganisms. And results are appropriate to improper method used.

So, yes. I'll point out it every time. No matter how many times you do it. Saying that you are comparing motion by taking static screenshots is as bad method as one can use.
Even worse when you claim that native is in motion thanks to you disabling AA.

It is ridiculous that you belive it is acceptable.

 

DLSS in CP2077 let’s my laptop 90W 2060 run absolute circles around my desktop 1080ti (and in quality/balance mode looks better than native).

 

GTFO for once again giving your deeply intimate opinion about things that not only you never used (like with DSLL), but in this case you never even heard of (SSIM)

I'll stick with raw numbers and objective image analysis.

You're hallucinating as usual.
I never said that A N Y of that. See why discussion with you is like doing chores

 

TL;DR: You are triggering yourself for sake of being triggered as you create situations which trigger you.
Why are you so triggered when someone objectively looks at available information. (I do hope you understand meaning of word "information".)
And then decides that it is not as fine and dandy as you see it?

And why are you so triggered at me? Just because I do not buy nVidia's GPUs when you have dozens of people with RTX cards on this forum who tell you same thing over time?
Because I have no problem to call your improper methods? And others mostly gave up?
- - - -
You could have noticed that I do not have problem with your opinion on tech or tech itself. I have problem on lack of objectiveness.

Spoiler: And it is still same problem (Paraphrasing):

Me: "DLSS 1.0 is good at AA, bad with artifacting at places."
You: "It is holy grail."
nVidia: *releases new version*
Everyone sane: "They improved DLSS, so it was likely not as perfect as some may think."
...
...
Me: "DLSS 1.9 is not there, yet."
You: "DLSS is best thing ever and 2.0 will prove you wrong."
Everyone sane: "That was not really an argument."
nVidia: *releases DLSS 2.0*
Me: "Improved on something, but has more artifacting and more blurriness."
You: "It is holy grail. It looks better than native."
nVidia: *Were still working on improvements."
...
...

You know, in those past 2 years, it is still same circle. And ultimately, it is nVidia which proves wrong those who believe that DLSS is already "Be-All & Is-All".
Except that objectivity is out of window completely, now. And You intentionally use lower quality rendering for "native" images, so you can claim that DLSS is closer to reference.
And you did not even write about method you obtained such reference.

At any point in time, you had option to actually give factual description on method which you used to obtain each of images. It was your sole decision to leave facts in obscurity.
And it was your decision to get angry when you use static viewport and scene as prove of temporal stability where you disabled TAA. (As someone pointed out it reduces image quality due since most implementations are blurry.)

And 1st thing you did with opposite opinion was that you called those people "sycophant = Shills who shill for some kind of gain".
So, why don't you clean the air by providing facts? You are in control of scenarios you presented. Nobody else changed settings in game for you, right?
And why don't you say who gets some kind of benefits from AMD or other party for opposing your opinions? And then deliver proof.

 

On the actual subject, I wish every single game came with DLSS. This plug-in will make a lot of ports so much easier. Anytime knows the effect engine version requirement it has?

 

Not sure, I just know it req's ue4 v4. 26 overall. This plug in is going to be a big deal for getting dlss into games... This was a big move.

 

UE4 seems to have a good upscaler also, but it's fairly new and most games don't use it, might as well do that too.

 

I'm happy that from all the engines it's UE,my favorite one visually.

 

In UE4 DLSS demos on YT, I did notice that reflections on objects had noticeably lower resolution than on native.
It almost felt like they were not being upscaled. Maybe it has something to do with problematic motion vectors with spherical reflections. (This certainly can be solved. But it may have some extra computational overhead depending on method which generates reflections. Like if actual object placement in reflection is raytraced, then raytracing may be required to get accurate size+placemnt of motion vectors.)

Then there can be pretty nasty problem in general for DLSS 2.0 due to need for motion vectors. Take surface which is 50% transparent and 50% reflective. There DLSS needs motion vectors and clean image of background seen through transparancy. And separate motion vectors and clean image of reflected objects.
And that's currently problem as games produce final composite image with motion vectors of objects in scene.

On such objects, I think upscalers will need some kind of logic, which will switch OFF temporal and use other method.

 

To be honest I haven't seen the code of an upscaler, so I don't really know. I don't expect it to work much differently from TAA, SMAA 2TX in that aspect. I think that past that stage is where the neutral network gets the image anyway, but I might be wrong.

 

Christ!

If I wanted to read madman's stream of consciousness, I'd read Virginia Wolf