• 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

sharknice

Supreme [H]ardness
Joined
Nov 12, 2012
Messages
4,883
I've been saying this is going to happen eventually. Well it's officially happening now.

DLSS5 uses generative AI to improve the graphics.


View: https://www.youtube.com/watch?v=jw_o0xr8MWU

1773686436032.png

1773686474691.png


1773686239379.png


1773686201726.png



The only thing they showed was some short video clips of it in different games, with no details of when it's going to be released, how they're doing it, or anything like that.
They'll probably release more information soon.

edit, it's here: https://www.nvidia.com/en-us/geforce/news/dlss5-breakthrough-in-visual-fidelity-for-games/
 
Last edited:
Looks like slop injected into games, the digital foundry video they seemed hyped for this. Personally, not a fan. Everything just kind winds up looking the same. All of the women look like they have lipstick on. If it was just a way of "fixing" lighting i'd be more into it. But the image of Grace from RE9 is awful.
 
Last edited:
Looks like slop injected into games, the digital foundry video they seemed hyped for this. Personally, not a fan. Everything just kind winds up looking the same. All of the women look like they have lipstick on. If it was just a way of "fixing" lighting i'd be more into it. But the image of Grace from RE9 is awful.
It makes current gen look like last-last-gen, but then again you can just not enable and stay stuck in the past.
 
Don't like it. Seems like it would change the artistic style too much. I'm okay with using AI to assist in making the 3D models or textures look more realistic in the development phase of character creation, but applying it over as a filter after is not a good idea. I am assuming it will be an option to disable, but don't like how it is being rolled into "DLSS". Should be called something else. DLSS should focus on better upscaling with less ghosting or other issues.
 
Last edited:
Don't like it. Seems like it would change the artistic style too much. I'm okay with using AI to assist in making the 3D models or textures look more realistic in the development phase of character creation, but applying it over as a filter after is not a good idea. I am assuming it will be an option to disable, but don't like how it is being rolled into "DLSS". Should be called something else. DLSS should focus on better upscalin with less ghosting or other issues.
I agree, fully. This needs a different name, and it changes the original style which is a no for me.

I don't mind if the devs use it, as long as they review the output and it matches what they want to achieve.
 
Last edited:
Don't like it. Seems like it would change the artistic style too much. I'm okay with using AI to assist in making the 3D models or textures look more realistic in the development phase of character creation, but applying it over as a filter after is not a good idea. I am assuming it will be an option to disable, but don't like how it is being rolled into "DLSS". Should be called something else. DLSS should focus on better upscalin with less ghosting or other issues.
Yeah that is my main concern, how will this affect different art styles? For typical AAA photo-realism slop this will actually be fantastic most of the time IMO. I don't see it working well for other types of games though, or being detrimental and homogenizing how games look.

Although I wonder how feasible it would be to have different models that are tuned for different art styles, like say a model tuned for Borderland's comic book aesthetic.
 
Although I wonder how feasible it would be to have different models that are tuned for different art styles, like say a model tuned for Borderland's comic book aesthetic.
that would be the way to go imo, studio opt-in and use render farm to provide a look and that what use (or small studio that want just photo realism of course those could work out of the box).

Right now dev can tune the lighting and other option, but you probably need training set to define your style correctly.
 
Last edited:
Looks like I'm waiting for DLSS 5 to play RE: Requiem - seeing as no GPU release this fall (RTX 6K series delayed right?) I'm assuming this means 5 series is supported?

DLSS 5 will come to games including AION 2, Assassin’s Creed Shadows, Black State, CINDER CITY, Delta Force, Hogwarts Legacy, Justice, NARAKA: BLADEPOINT, NTE: Neverness to Everness, Phantom Blade Zero, Resident Evil™ Requiem, Sea of Remnants, Starfield, The Elder Scrolls IV: Oblivion Remastered, Where Winds Meet, and more.

DLSS 5 will arrive this Fall. See the first preview this week at GTC, and below, you can check out several examples of NVIDIA DLSS 5 in Resident Evil™ Requiem, EA SPORTS FC™, Starfield, Hogwarts Legacy, and the NVIDIA Zorah tech demo.
 
I'm curious what kind of horsepower it's going to require. I like the visuals, but am I gonna need a 7090 to take advantage?
right now they are running this either with a 5090 just for dlss 5 or using large vram (RTX 6000 i imagine) to do it, but they expect to be running on regular GPUs soon enough. according to DF video.
 
Yeah that is my main concern, how will this affect different art styles? For typical AAA photo-realism slop this will actually be fantastic most of the time IMO. I don't see it working well for other types of games though, or being detrimental and homogenizing how games look.

Although I wonder how feasible it would be to have different models that are tuned for different art styles, like say a model tuned for Borderland's comic book aesthetic.

I'm sure that will come. We'll also eventually see players able to set it to what they want with a drop down of options, and even just full control with a prompt.
 
I'm sorry but it just gave Grace AI filter face. It looks like shit tbh. If it can just improve lighting quality without touching the original intent of the model that's one thing. But I'd rather have worse lighting and actual models with their original textures, not slop face on everyone.

So don't use it, turn it off in your games - I like it from what I see I'll (presumably) leave it on 🤷‍♂️
 


Says in the video at 4:20 they're doing it on 1 GPU currently in the lab (2 GPU was to make sure demo went smoothly) with non-specific 'not so high VRAM requirements'

I won't lol. Kind of sad people want to ruin works of art with a trash filter. Not unsurprising though

Yeah great thing about playing games is people get to play them how they want to play them not how you say they should play them - stay mad though? 🤷‍♂️
 
This is going to open a door that many have been trying hard to keep shut. I think the reason it's taken this long is because of same face. If they've figure out how to overcome same face, then subjectively it will win over most people.
 
1 GPU currently in the lab (2 GPU was to make sure demo went smoothly) with non-specific 'not so high VRAM requirements'
Geforce Now having 48GB vram vGPUs per session available could maybe a first way to make this broadly available (all that vram is a bit wasted right now).
 
Don't care.

Sidenote: It's funny how Nvidia plows ahead blindly while AMD just takes notes and applies features pragmatically.

This right here is why I am on the Red Team, far less risky.
They'll wind up doing the same garbage. This is the first shot across the bow of the games industry generally IMO. Why would you want to work as an artist for all of your art to be put behind a slop filter. Just one step closer to games being AI generated trash.
Says in the video at 4:20 they're doing it on 1 GPU currently in the lab (2 GPU was to make sure demo went smoothly) with non-specific 'not so high VRAM requirements'



Yeah great thing about playing games is people get to play them how they want to play them not how you say they should play them - stay mad though? 🤷‍♂️
Not mad, just disappointed, but again, not surprised this is what people want. My guess is it will only be availble for the RTX 6000 series.
 
My guess is it will only be availble for the RTX 6000 series.

They said DLSS 5 is gonna be released in fall - thought RTX 6K series is delayed so not coming out in fall? Unless they're gonna surprise everyone with that this fall too (doubt it though personally, just based on current HW/RAM/etc landscape) 🤔
 
Apparently based on DF if peeps don't want to watch the vid, it's running on two 5090's atm. One handling DLSS, the other handing the game based on what DF stated. Honestly as long as you can toggle it separately, I think that's fine...But as someone getting into game dev, I hate it from an art control and intent standpoint. Sacrificing the art direction for an ai face filter is just annoying that we are heading in this direction. I'll admit it's better than I expected, and what we've seen in the past...But this is a very fine line that is being stepped over. Not sure how I feel atm to be honest. Time will tell.

As nice as the lightning looks for environment. A bit jarring mixing in realistic lighting and CG content. The contrast between the two feels very awkward to look at...Like that birds jungle scene. Very weird....
 
Why would you want to work as an artist for all of your art to be put behind a slop filter.
Right now you can define look via sementic like text, this is blue with a bit of transpiration feel on the material or blade runner atmosphere to the room) and so on and more importantly they can just say to the AI to not (scoring how much) to change things at all per section, it include intensity and masking lever.

Ideally they would be able to make a training set with their look and used it as a local adaptar/neural material descriptor over DLSS 5 work, I think that already what it will be at launch. And it will get multimodal, artist will be able to not just use text but a painting they do, music piece, picture, motion capture of an actor and so on.

For small dev that could be incredible, you add in your game engine make that lake a bit foggy/swamp look and so on, but keep my black and white Hitchock ligthing look instead of color photorealistic, look what it look like and play with intensity levels.
 
Last edited:
They said DLSS 5 is gonna be released in fall - thought RTX 6K series is delayed so not coming out in fall? Unless they're gonna surprise everyone with that this fall too (doubt it though personally, just based on current HW/RAM/etc landscape) 🤔
Interesting, they said they had to use two 5090s to run it currently. I wonder if VRAM is a limiting factor?
 
Interesting, they said they had to use two 5090s to run it currently. I wonder if VRAM is a limiting factor?
they pointed that it would vram issues in DF video, as it was already running on a single one at Nvidia, if it is not on a kept secret Rubin gaming GPU...nothing is faster than a 5090 at this that can play game?, probably RTX 6000 to allow the vram usage.

If it is like other ML model, you do a big model, then you do a distilled and quantized down version (gemini flash of gemini, etc...) that give you 80-90% of the capacity but take way less vram and compute once it work well to be the workhorse, that could be why they are confident to be able to size this down.
 
Says in the video at 4:20 they're doing it on 1 GPU currently in the lab (2 GPU was to make sure demo went smoothly) with non-specific 'not so high VRAM requirements'



Yeah great thing about playing games is people get to play them how they want to play them not how you say they should play them - stay mad though? 🤷‍♂️
Also guessed that it might be coming for cheap crummy consumer gaming GPUs, like the 5090 (though, you might need 2).
 
This is just going to make all games have that same AI filter look on top. I'd rather have the original intended artstyle/designs of the devs. I wouldn't mind using it just for better environmental details/lighting though.
 
  • Like
Reactions: erek
like this
This is just going to make all games have that same AI filter look on top. I'd rather have the original intended artstyle/designs of the devs. I wouldn't mind using it just for better environmental details/lighting though.
That's kind of where I'm at with it, if I want the hyper real light filter with no material changes that's cool as an option.

As long as it stays entirely optional I don't care if the slop filter is there ultimately. If it's required in the future, I'm not playing your game that looks like AI realism.
 
  • Like
Reactions: cvinh
like this
Yeah but we don't know what GPU they're using "in the lab" it might be something not released yet. I have no idea. Again I don't really care I just won't use it if I switch to an Nvidia GPU in the future.

Think it's safe to assume if using 2x 5090s in the demo then in the lab it's 1x 5090 - they explained in the video they were running it on 1x 5090 (generative part alone) in the demo and the other 5090 was there to run everything else (like upscale and framegen?) so it all went smooth - while in the lab it's all in total running on 1x GPU as mentioned
 
Back
Top