• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

DLSS 5 - Generative AI

1788108062353.png


M76 said:
Filling in the gap between two pixels = interpolation. Completely making up (Guessing) the details is generative AI. They are fundamentally different. It's neither a good nor a bad thing in of itself, so IDK why deny it.
Click to expand...

Filling in the gap between two pixels = generating the information for that missing gap - the distinction you are trying to make is simply a colloquial and marketing one, not a technical nor practical one. They're all generative AI. DLSS5 isn't making stuff up out of thin air, it's referencing and inferring things from preexisting game data like vectors and textures and frames too just as upscale and framegen do.

Edit: Part of the appeal even of DLSS upscale is that it can GENERATE details in the higher res output that should/would be there if native (or even that should be there, and NOT in native), that are NOT in the lower res input - people have shown examples of this on this very forum with images containing wires and poles and how with DLSS upscale they look even better than native etc
 
Last edited:
kram182 said:
Edit: Part of the appeal even of DLSS upscale is that it can GENERATE details in the higher res output that should/would be there if native (or even that should be there, and NOT in native), that are NOT in the lower res input - people have shown examples of this on this very forum with images containing wires and poles and how with DLSS upscale they look even better than native etc
Click to expand...

Indeed!

1788115671015.png


Bear in mind, this freaking DLSS 2.3 lmao. The super ancient CNN model of version 2.3 already looks better than native + TAA. Yet you have some people coping that native is still better. I mean native + DLAA yeah sure obviously that's better, but native resolution with TAA or no AA has always looked worst to me than DLSS.
 
MistaSparkul said:
Indeed!

View attachment 822913

Bear in mind, this freaking DLSS 2.3 lmao. The super ancient CNN model of version 2.3 already looks better than native + TAA. Yet you have some people coping that native is still better. I mean native + DLAA yeah sure obviously that's better, but native resolution with TAA or no AA has always looked worst to me than DLSS.
Click to expand...

And another way we can put it - if that's just 'reconstruction' - then DLSS5 is merely 'reconstructing' shadows, ambient occlusion, subsurface scattering, highlights, reflections, etc etc
 
M76 said:
So literally not generative AI. There is no need to fanboy so hard that you misrepresent things. No one would normally call a machine learning based upscaler generative AI. DLSS pre 5 is an upscaler / frame interpolator nothing more, it does not try to make up details that were never there originally, it tries to fill the missing data between two existing pixels / frames. A wholly different concept than DLSS5 which is putting a completely new facade on things. Whether you like the results or not there a significant difference. I'm not against it, but not acknowledging this fundamental difference just gives ammo to the detractors.
Click to expand...
Not sure here, the "16K image" it tried to generate never existed (the 4k native version of the game neither for all the model know)

DLSS 4.5 transformer is generative AI, it "recognise" that something is text, foliage, hair, etc... and generate what it estimate should be there and generate structural details, that why it sometime hallucinate
 
The creative director of Kingdom Come 2 has some chosen words:
https://x.com/DanielVavra/status/2094125947515478452?s=20
I repeat: It doesn’t alter the geometry of the models or the appearance of the characters in any way!What it does is take existing normal maps and create much better skin texture and facial detail (examples 6 and 2) than the lighting model in the game. It also looks a bit like an attempt at subsurface scattering and other advanced skin rendering techniques. But sceptics could say, that its just a fancy version of clarity and dehaze filters which have very similar effect. I often criticized the character artists’ work when certain faces in the game looked lacking in detail and plasticity, and they’d often show me that it looked fine on their computers and that the problem was with the lighting in the game. Well, now it looks right in the game, too! This is how those faces were supposed to look. The AI isn’t making anything up.Above all, though, it makes ambient occlusion look like it’s on steroids. You have no idea how terribly—and I mean really, really—annoyed I was that in KCD2 we couldn’t render shadows from hats, hoods, and helmets on characters’ heads, or even shadows from small details on character models (buckles, laces, pouches). DLSS5 does this, and it does it absolutely brilliantly! (see examples 1, 3, 7, 8, and 11). On top of that, it improves the rendering of shadows in hair (5), so they finally have the right density and darkness.It doesn’t do much in the environment, but interestingly, it lightens shadows in some places and renders textures within them, while darkening them where it makes sense (10, 12, 13, 14). So there’s a slight improvement here as well. It also renders shadows in grass and vegetation better. Which is definitely a good thing.For me, this is primarily an excellent photo realistic upgrade to the character lighting model.
Click to expand...
 
littlegreenmenfromhome said:
The creative director of Kingdom Come 2 has some chosen words:
https://x.com/DanielVavra/status/2094125947515478452?s=20
Click to expand...

It is absolutely hilarious to me that people want to defend "original artistic vision" when the original vision, as he says, lacks detail and has a "plastic" look to faces with current lighting models. DLSS5 is literally making the original artistic vision look closer to intended by fixing the lighting problems lmao. Haters can COPE MORE.
 
MistaSparkul said:
It is absolutely hilarious to me that people want to defend "original artistic vision" when the original vision, as he says, lacks detail and has a "plastic" look to faces with current lighting models. DLSS5 is literally making the original artistic vision look closer to intended by fixing the lighting problems lmao. Haters can COPE MORE.
Click to expand...
This is a dumb take.
 
Shoganai said:
This is a dumb take.
Click to expand...

Care to explain why?

"I often criticized the character artists’ work when certain faces in the game looked lacking in detail and plasticity, and they’d often show me that it looked fine on their computers and that the problem was with the lighting in the game. Well, now it looks right in the game, too! This is how those faces were supposed to look. The AI isn’t making anything up"
1788122793006.png


If you know more than the literal creative director of the game, I'm all ears. Otherwise if you got nothing, then you're the one with the dumb take.
 
Last edited:
MistaSparkul said:
If you know more than the literal creative director of the game, I'm all ears.
Click to expand...
Also to anyone who is arguing "creative intent": You'd better have never used a reshade filter, or texture pack, or any kind of mod really or you are altering the creative vision of the artists/developers! :LOL:

I think it is funny how suddenly a ton of people supposedly give a shit about the vision/intent of the developers with regards to DLSS 5, when we never heard from them before on any of the other technologies that are used to alter the way a game looks or plays. Now they care so much they'll defend it from the very people who created the game! "No, you don't understand! This isn't the vision you had, it is bad and evil!"

It is just hate for hate sake.


The real issue, which he mentions, is the performance cost. We'll see if they can improve it any, and maybe games that use the filters to only use it on certain things won't get hit as hard but if not, it'll be one of those that you'll really have to think about. Like I can't imagine using it in Cyberpunk, already that's running at only around 50-70fps native with path tracing on. I'm not going to turn PT off, and I am not willing to cut that in half so I can't see using it there.
 
Sycraft said:
Also to anyone who is arguing "creative intent": You'd better have never used a reshade filter, or texture pack, or any kind of mod really or you are altering the creative vision of the artists/developers! :LOL:

I think it is funny how suddenly a ton of people supposedly give a shit about the vision/intent of the developers with regards to DLSS 5, when we never heard from them before on any of the other technologies that are used to alter the way a game looks or plays. Now they care so much they'll defend it from the very people who created the game! "No, you don't understand! This isn't the vision you had, it is bad and evil!"

It is just hate for hate sake.


The real issue, which he mentions, is the performance cost. We'll see if they can improve it any, and maybe games that use the filters to only use it on certain things won't get hit as hard but if not, it'll be one of those that you'll really have to think about. Like I can't imagine using it in Cyberpunk, already that's running at only around 50-70fps native with path tracing on. I'm not going to turn PT off, and I am not willing to cut that in half so I can't see using it there.
Click to expand...

Wouldn't you just turn PT off and use DLSS5 to handle all the lighting? Stacking PT with DLSS5 seems unnecessary. At least with games that lack any PT at all, the performance hit using DLSS5 would be no worst than if the game did have PT and you were only using that. KCD2 is a good example of that, game runs pretty well but lacks really good lighting and has no PT option, you can kick on DLSS5 and combined with SR the frame rate would be quit acceptable.
 
Last edited:
MistaSparkul said:
Indeed!

View attachment 822913

Bear in mind, this freaking DLSS 2.3 lmao. The super ancient CNN model of version 2.3 already looks better than native + TAA. Yet you have some people coping that native is still better. I mean native + DLAA yeah sure obviously that's better, but native resolution with TAA or no AA has always looked worst to me than DLSS.
Click to expand...

It depends a lot on the game. Shimmering and whatnot was and still is common. With the latest version of DLSS in Death Stranding 2 I found the shimmering on bushes to be too distracting. The in engine version looked a bit better but still looked bad. I decided to turn off DLSS and turn on frame generation. Problem is I ran out of VRAM with ray tracing and frame generation which resulted in stutters. So I turned off ray tracing. 16GB is not enough VRAM for some games at 1440 if you want to use ray tracing and frame generation.
 
MistaSparkul said:
Wouldn't you just turn PT off and use DLSS5 to handle all the lighting? Stacking PT with DLSS5 seems unnecessary. At least with games that lack any PT at all, the performance hit using DLSS5 would be no worst than if the game did have PT and you were only using that. KCD2 is a good example of that, game runs pretty well but lacks really good lighting and has no PT option, you can kick on DLSS5 and combined with SR the frame rate would be quit acceptable.
Click to expand...
I mean, we'll see and I'll test it, but I suspect not. Path tracing does a whole lot more than I think DLSS5 is capable of. One simple but good example would be off-screen reflections. If there is an object out of screen space that should be reflected/refracted in a surface, path tracing can do that. There's no way for DLSS5 to do it, since it it inference based on the final scene.

It is going to be something I'm going to play with a lot for sure. I'm most interested to see it implemented in games, by the artists, and see how it looks there. I think when someone has fine-grained control over it, where it is used and when, and also changing what filter is used and how much scene-by-scene it'll be the best. However I'm for sure going to try stacking it on existing games and seeing what happens. I suspect in many cases the juice isn't worth the squeeze for me at this point since it looks like an improvement, not a transformation, and there's only so much I'm willing to pay for that.

HOWEVER I haven't seen it in game, in motion yet. Maybe it hits harder there. Path tracing is like that for me. Like it looks good in screen shots, and better in video, but to really appreciate it, I had to experience it in game. It is the way everything looks so... right that makes it transformative to me. Light, shadows, reflections, etc just interact in a way that feels like they should. If DLSS 5 is similar, I'll probably use it a lot.

Also depends on the game and how much latency bothers me and how well it works with framegen. I'm really not willing to go much below 50-60 rendered FPS. Past that, things start to feel too choppy to me even when FG is working well. So if it pushes a title down in that range, probably won't use it. Likewise I really like a display framerate of at least 90fps, preferably more, so if FG doesn't work well with a game, that is another issue because if the game runs at 100fps without DLSS 5, 50fps with it and FG has issues, I'm going to choose 100 without personally.

I'm excited though. I plan on playing with it soon it is just... Ummm... I might be spending all my time playing Everquest Legends right now, and that game has graphics that no filter can save :D
 
MistaSparkul said:
View attachment 822981
Click to expand...

"Warhorse Developer Defends DLSS 5, Says It's Not an AI Slop Filter​

by DrUsmanSaleem Today, 20:36 Discuss (2 Comments)
Kingdom Come: Deliverance 2 director Daniel Vávra has praised DLSS 5 again, saying the tech now looks closer to how his team originally envisioned the game's lighting and character faces. Speaking on X, he said character models that once looked flat in-engine finally read the way artists intended, closing the gap between how faces looked on artists' workstations and what actually shipped in the game.

Vávra singled out shadow rendering as a big win. He said Warhorse struggled for years to get shadows from hats, hoods, helmets, and small props like buckles and pouches to render on characters, calling it a persistent frustration during development. DLSS 5 apparently solves this, and Vávra called it an excellent upgrade to the character lighting model, though he noted it roughly halves frame rate in its current form. He also sees potential in DLSS 5 when applied to older games:
"It's not the infamous AI slop we saw couple months ago. I see huge potential here for the future, and it's absolutely great that even older games lighting could be improved this way.""
Click to expand...
 
kram182 said:

View: https://www.youtube.com/watch?v=sw12oLTRH38
Click to expand...

The opening scene there is interesting because it does the opposite with Grace of what people were going on about: Rather than making her look like she is all dolled up wearing makeup, it makes her look way more tired, disheveled, and stressed. In the 4.5 shot she looks like she has a very smooth, perfect, layer of makeup on (or more realistically, just less detail because, you know, video game rendering).
 
It's amazing what it's capable of when you put it in the hands of gamers, instead of tech execs who want to push AI slop.
 
Sycraft said:
Also to anyone who is arguing "creative intent": You'd better have never used a reshade filter, or texture pack, or any kind of mod really or you are altering the creative vision of the artists/developers! :LOL:

I think it is funny how suddenly a ton of people supposedly give a shit about the vision/intent of the developers with regards to DLSS 5, when we never heard from them before on any of the other technologies that are used to alter the way a game looks or plays. Now they care so much they'll defend it from the very people who created the game! "No, you don't understand! This isn't the vision you had, it is bad and evil!"

It is just hate for hate sake.


The real issue, which he mentions, is the performance cost. We'll see if they can improve it any, and maybe games that use the filters to only use it on certain things won't get hit as hard but if not, it'll be one of those that you'll really have to think about. Like I can't imagine using it in Cyberpunk, already that's running at only around 50-70fps native with path tracing on. I'm not going to turn PT off, and I am not willing to cut that in half so I can't see using it there.
Click to expand...
DLSS 5 right now doesn't support upscaling, it is native or DLAA.
I suspect that will be ixed, remeber all this is a "beta" .dll hacked to run in other games.
 
erek said:

"Warhorse Developer Defends DLSS 5, Says It's Not an AI Slop Filter​

by DrUsmanSaleem Today, 20:36 Discuss (2 Comments)
Kingdom Come: Deliverance 2 director Daniel Vávra has praised DLSS 5 again, saying the tech now looks closer to how his team originally envisioned the game's lighting and character faces. Speaking on X, he said character models that once looked flat in-engine finally read the way artists intended, closing the gap between how faces looked on artists' workstations and what actually shipped in the game.

Vávra singled out shadow rendering as a big win. He said Warhorse struggled for years to get shadows from hats, hoods, helmets, and small props like buckles and pouches to render on characters, calling it a persistent frustration during development. DLSS 5 apparently solves this, and Vávra called it an excellent upgrade to the character lighting model, though he noted it roughly halves frame rate in its current form. He also sees potential in DLSS 5 when applied to older games:
Click to expand...
So much for "A.I. Slop", eh?
 
littlegreenmenfromhome said:
DLSS 5 right now doesn't support upscaling, it is native or DLAA.
I suspect that will be ixed, remeber all this is a "beta" .dll hacked to run in other games.
Click to expand...

Even if you click on "Enable Upscaling", it does nothing? I saw one example where clicking on that clearly made the image lower res.


View: https://www.youtube.com/watch?v=_93p1IMHnVA

He enables upscaling in the beginning of the video and the image got more shimmering.
 
littlegreenmenfromhome said:
DLSS 5 right now doesn't support upscaling, it is native or DLAA.
I suspect that will be ixed, remeber all this is a "beta" .dll hacked to run in other games.
Click to expand...
So that appears to not be accurate.

I've tested DLSS 5 on Satisfactory and it looks like upscaling is working fine. I think what people are talking about is that DLSS 5 is applied AFTER the upscaling. So whatever scaling factor you have for rendering, you then have to do DLSS-NR on your native rez image which if you are a 4k boi, means a lot of calculations. It looks like nVidia has been playing with the idea of trying to do it before, or as a part of, upscaling but that isn't how it works now.

We'll see what happens as it is released. Maybe that is one of those things that is almost working and will come out, or maybe it is something they have been playing with but looks like garbage and is nowhere near done and the release will be as it is now. Hard to say.

But as it stands upscaling DOES seem to work and will increase your FPS, but then the neural reconstruction part hits the final image and tanks your FPS because it has to be processed at native rez. I do imagine that turning down resolution will help its speed tremendously.


As for testing it, Satisfactory is probably one I'll pass on using it in. I'll play with it more because of course I will, but initial impression is that it does look nice, and improves some things like material textures, lighting under a shadow'd surface, and so on but it is a fairly minor improvement, not worth the 50% (and even more sometimes) FPS hit. Interestingly it seems to clean up some of the Lumen noise. Satisfactory doesn't implement Lumen very well, it wasn't designed for it and they just basically added a switch to turn it on and said "Use it if you want to," so it isn't optimized well and there's a ton of noise on many surfaces. DLSS-NR reduces it. Doesn't eliminate it, but it reduces it.
 
Technology can’t be stopped, especially the kind that most people like.

If DLSS5 is accepted as a “modification” (as is currently the case), then people will like it even more.

And Nvidia is very good at manipulation.
When they announced DLSS5, everyone pounced on them with talk of “AI slop,” so Nvidia let it “leak,” and as we know, when something is pirated, it’s great because it’s free, and people like it precisely because it’s stolen.

The filter works not bad—yes, it distorts the light source, but who cares when things look “better than native”...
I just want to mention this for everyone who’s now happy that this is just another nail in the coffin of gaming.
 
Sycraft said:
How is better technology a "nail in the coffin" for gaming?

God people can be melodramatic.
Click to expand...
There has been a war against gaming studios for several years now, aimed at bringing them under control, and so far, they’re losing it.
Nvidia is part of this and serves as a tool for that purpose.
But don’t worry, everything’s fine—be happy.
 
MistaSparkul said:
Even if you click on "Enable Upscaling", it does nothing? I saw one example where clicking on that clearly made the image lower res.
Click to expand...
From my understanding it will use regular DLSS 4.5 to do the upscaling and DLSS 5 will work from the final render at the very end of the pipeline.... it is not the DLSS 5 NR model that do it in one go. It does not upscale yet.

Sycraft said:
As for testing it, Satisfactory is probably one I'll pass on using it in. I'll play with it more because of course I will, but initial impression is that it does look nice, and improves some things like material textures, lighting under a shadow'd surface, and so on but it is a fairly minor improvement, not worth the 50% (and even more sometimes) FPS hit. Interestingly it seems to clean up some of the Lumen noise. Satisfactory doesn't implement Lumen very well, it wasn't designed for it and they just basically added a switch to turn it on and said "Use it if you want to," so it isn't optimized well and there's a ton of noise on many surfaces. DLSS-NR reduces it. Doesn't eliminate it, but it reduces it.
Click to expand...
We will have to see if it is not worth that 50% fps hit.... i.e. some games (like Ubisoft engine) at low setting + DLSS5 seem to look better than at ultra setting without it and the performance hit being almost nill, we can imagine people finding what it is really good at reconstructing and whichs setting you can go to low/medium without loosing much quality and what you need to keep, so the general fps hit will not be that harsh
 
mullet said:
I can't imagine what the DLSS8 thread will look like.
Click to expand...
Honestly I am pleasantly surprised with the takes here. This is an older forum, so I figured most people here would be anti-AI anything. Cool to see lots of reasonable takes with nuance here. Better then what I'm seeing from my peers in discord.
 
AVATARAT said:
The filter works not bad—yes,
Click to expand...
I think that a general misunderstanding of how different this is, than your traditional sharpening or others filters.

This has the ability to add new information to the signal, filtres apply deterministic mathematical operation to a signal and are fundamentally destructive (we remove noise and other issues to it when you make a filter, you do not add).

This is more analogue during a film restoration process of the steps when human artist painted on the frame (litteraly or using a computer), using human intuition to add to the image missing information.

This is a generative synthesis step rather than a transformative filter, it "recognize" that this is skin and next to it its metal and having seen what it look like in extreme high quality, having access to depth map and other input hallucinate what it should look like.
 
ivandagiant said:
Honestly I am pleasantly surprised with the takes here. This is an older forum, so I figured most people here would be anti-AI anything. Cool to see lots of reasonable takes with nuance here. Better then what I'm seeing from my peers in discord.
Click to expand...
I think there is 2 different kernel of issues, AI in particular but an older and more general closed API (would it be DirectX by microsoft, Glide by 3dfx and so on) with hardware vendor.

This I would imagine, everyone can clearly see how giant potential it has and that part of the issue even, because if game does not look the same depending on which vendor you run the hardware from (or do not run at all) that as its own list of possible issues that totally different than AI.

The moment this is running via an NPU with sram inference running almost instantly, that the game engine take advantage and that the training dataset contains this specific game engine data (i.e. it create a giant bunch of data where the engine created basic real time info and super computer movie quality renders for each frame), this could let mobile Nintendo Switch 3 level hardware of 2032 running in real time early 2000s Pixar/Dreamwork level of quality...

If the model are trained by the game maker and run on any hardware, that could change things a lot for a lot of people, it would become just modern rendering pipeline that some use, being Nvidia own models that are being used at the moment feel different.
 
Last edited:
MistaSparkul said:
Indeed!

View attachment 822913

Bear in mind, this freaking DLSS 2.3 lmao. The super ancient CNN model of version 2.3 already looks better than native + TAA. Yet you have some people coping that native is still better. I mean native + DLAA yeah sure obviously that's better, but native resolution with TAA or no AA has always looked worst to me than DLSS.
Click to expand...

I mean if you want to get down to the nuts and bolts of it, I don't understand how TAA is more "native" than DLSS in the sense that TAA is trying to smooth out the image by comparing it to previous frames. In both cases you're using sampling data to generate what the program decides the image should look like, only one looks like ass and the other doesn't. If you want to get technical, neither is a "native" image. The native image has aliasing. You're applying an additional layer to try to smooth that out, modifying the "native" image to make it look better, and both come at some sort of compute cost. Just use whatever you think looks good and move on with your life. I don't know what people reflexively have to hate something just because it's "AI" or "AI adjacent".
 
kram182 said:
Filling in the gap between two pixels = generating the information for that missing gap - the distinction you are trying to make is simply a colloquial and marketing one, not a technical nor practical one. They're all generative AI. DLSS5 isn't making stuff up out of thin air, it's referencing and inferring things from preexisting game data like vectors and textures and frames too just as upscale and framegen do.
Click to expand...
The literal definition of generative AI that it generates new data. Not filling gaps in existing data. By your definition an OCR software would be generative AI because it fills in the characters based on partial data.

DLSS5 is image to image generative AI. It uses a clip like model to make its own prompts, the prompting strength driven by the parameter sliders it has. DLSS upscaling and framegen does no such thing, it has no prompts, it is not generative AI. IDK what is it you are trying to achieve by blurring the line between them. Trying to make it more acceptable? But genAI haters won't care if you lump in DLSS upscale into the genai bucket, they'd just say then we hate that too.
kram182 said:
Edit: Part of the appeal even of DLSS upscale is that it can GENERATE details in the higher res output that should/would be there if native (or even that should be there, and NOT in native), that are NOT in the lower res input - people have shown examples of this on this very forum with images containing wires and poles and how with DLSS upscale they look even better than native etc
Click to expand...
DLSS upscale looking better than native is nonsense, it is only perceived that way by some because it blurs edges acting as "free" anti-aliasing.
 
M76 said:
The literal definition of generative AI that it generates new data. Not filling gaps in existing data. By your definition an OCR software would be generative AI because it fills in the characters based on partial data.
Click to expand...

If there was data then there'd be no gap, there can only be a gap by definition with an absence of data/no data - that new/'non-existing before' data then 'fills' - whether missing data in-between frames or more/missing pixels/missing pixel data.

It's generative.

M76 said:
DLSS upscale looking better than native is nonsense, it is only perceived that way by some because it blurs edges acting as "free" anti-aliasing.
Click to expand...

It's been shown plenty of times, plenty of places - cope - if it's """perceived"""" to look better, then it *looks* better.

M76 said:
IDK what is it you are trying to achieve by blurring the line between them. Trying to make it more acceptable? But genAI haters won't care if you lump in DLSS upscale into the genai bucket, they'd just say then we hate that too.
Click to expand...

You're the one with an admitted and apparent AI-hating agenda here, I'm just explaining things. You can never turn DLSS 5 on for all I give a damn.
 
Last edited:
M76 said:
The literal definition of generative AI that it generates new data. Not filling gaps in existing data. By your definition an OCR software would be generative AI because it fills in the characters based on partial data.
Click to expand...
yes which DLSS 4.5 transformer do, it recognize that something is hair, text, cable and generate new data to fill it (it never seen the actual 4k image or the 16k training version, that never exist here), it is not working with base resolution and the 16k to fill gaps.

M76 said:
DLSS upscale looking better than native is nonsense, it is only perceived that way by some because it blurs edges acting as "free" anti-aliasing.
Click to expand...
Looking better is quite subjective, but objectively it does add missing information into the image, that became less than 1 pixel in the low bit float precision render pipeline and not there.

A good example would be the text you just talked about, once it reconize that it is a stop sign it is relatively easy to understand how it would be possible for it to look better than native, the same apply to everything just in less straitforward ways.
 
Last edited:

“Leaked DLSS 5 is Being Tested on RTX 20-series "Turing" GPUs, Now Runs on Emulators and Old DirectX Titles​

by DrUsmanSaleem Today, 10:00 Discuss (2 Comments)
The leaked DLSS 5 DLL keeps finding new homes, and support has now spread to RTX 20-series "Turing" GPUs. That architecture is three whole generations behind the NVIDIA-recommended "Blackwell," and modders are showing no signs of stopping either. Developer ShortFuse has released new DLSS-NR builds for older RTX hardware, including one described as a "speed boost" build for RTX 20-series. The results are still a mixed bag, however. Some games run, while others crash or fail to initialize. Red Dead Redemption reportedly worked, while Red Dead Redemption 2 did not, and Hogwarts Legacy got stuck in a standby state instead of activating the feature. GTA 5 ran for about ten seconds before the display signal dropped out entirely.

Moreover, X user @DystopianSuns says the community now has DLSS 5 running on RTX 20 and RTX 30 GPUs using an FP16 implementation. This is a change from the FP8 approach that previously tanked performance on Ampere cards. FP16 should suit Turing and Ampere better, since neither generation accelerates FP8 the way Blackwell does. With this approach, users reported respectable performance in many titles on the RenoDX Discord using RTX 30-series cards. The same X user apparently got Manhunt running through the PS2 emulator PCSX2 with DLSS 5 enabled, and other testers report the mod working in older titles built on DirectX 9 and DirectX 11, well outside the APIs NVIDIA has confirmed for the feature.“
 
You must log in or register to reply here.
Back
Top