• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Nvidia at Live GTC : DLSS 5

Hopefully we can adjust the level of AI-ification.
I would say let us wait and see what they will achieve with this.
Raytracing was a joke/niche feature too and now it is able to transform the games.

What I dont like:

2 5090s were needed?! wtf...

What we definitely can expect ... AMD will follow 2-3 years later :D
 
People not excited about this are stupid or aren't dreaming big enough

This is amazing. Devs will be able to fine tune this, so they could make it work only with environments, or change character models to look good while using this tech etc.

This is like any other use of AI: can be good when used properly.

Two 5090s tho is scary tho. They have to optimize A LOT to make it usable.
 
Last edited:
KeplerL2 said:
AI slop filter with zero respect for artists intent and vision 🤢🤮
Click to expand...

It's in their FAQ. All of these results were created by these same artists.

How does DLSS 5 ensure image quality is consistent with the artist's intent?

DLSS 5 honors artistic intent in two ways:

  • Inputting the game's color and motion vectors for each frame into the model, anchoring the output in the source 3D content.
  • By providing developers with detailed controls such as intensity and color grading. Artists can use these controls to adjust blending, contrast, saturation, and gamma, and determine where and how enhancements are applied to maintain the game's unique aesthetic. Developers can also mask specific objects or areas to be excluded from enhancement.
 
Not only that it can be controlled, and it will by the competent, otherwise who cares about it anyhow. But look at what it does to Starfield, a shitty looking, low effort trash? It's as if they can do a half assed, low effort model, or like a concept, and AI simply finishes the job. IDK but i'm hyped for something after a long long time. DF says Todd is on it as well, haha.
 
Last edited:
Gonzito said:
So this wont be available for my 4090?
Click to expand...

They won't say. I kinda doubt 24gig VRAM will be enough to do this. Also bet this is gonna require multi frame gen.

I doubt this tech is ready for prime time until DLSS6 and new cards. Running an LLM and playing a game at the same time is too demanding.

Next cards are gonna be insanely expensive if they are committed to this because they're gonna need crazy amounts of VRAM.
 
peish said:
dlss-5.png
dlss-4.png
Click to expand...
Chud filter...I guess this is a win?
 
Wolzard said:
It's in their FAQ. All of these results were created by these same artists.

How does DLSS 5 ensure image quality is consistent with the artist's intent?

DLSS 5 honors artistic intent in two ways:

  • Inputting the game's color and motion vectors for each frame into the model, anchoring the output in the source 3D content.
  • By providing developers with detailed controls such as intensity and color grading. Artists can use these controls to adjust blending, contrast, saturation, and gamma, and determine where and how enhancements are applied to maintain the game's unique aesthetic. Developers can also mask specific objects or areas to be excluded from enhancement.
Click to expand...

Just because it says that, doesnt mean that it doesnt look like absolute shit compared to the original art????

You can clearly see a difference right.
 
If there's some way you can control it for just the lighting & environment it's neat.

Grace and the Starfield characters literally look like those caked up AI character revisions on X.
 
Last edited:
I'm a big fan of NVIDIA, but drawing the line here. These faces on DLSS 5 just look like someone used AI art to make a semi-realistic, but obviously AI generated person, then used that face as some sort of camera filter. I don't like it.
 
Well clearly it continues to be very divisive tech, just the same how the argument for native vs upscaled comparisons started initially.
What is shown here is both very impressive in some aspect, and clearly gonna need the time to deliver properly.
 
People being afraid of "artistic integrity" need to understand that DLSS isn't just something a developer implements and it works. Graphics need to be designed around specific functions otherwise they don't work or break a game. The "AI enhancements" on their faces aren't just generated out of nowhere with a prompt. It wouldn't be consistent this way.

You can say that the artistic vision of the developers is that transformed look.

The issue here is that it's uncanny. It needs animations to match the look.
 
LiquidMetal14 said:
I don't think people quite understand how this could potentially work to benefit and save on rendering time. People want shorter game cycles and lesser budgets and don't see the little ways that we can get it to be beneficial. Makes you wonder on this enthusiast gaming forum, you would think there would be more people that understand some fundamentals about this technology evolving. And that isn't some cynical Nvidia torch bearer type stuff.

This could really benefit a lot of little things like that. And the cool part is you can probably turn it off or cycle between different things. Again, I'm not sure why the outrage about some things but not yourselves out. Those of us that care about tech and all these little things that are advancing will gladly take part in this and if we don't like it we will disable it. Novel concept.
Click to expand...
Plus you will be able to adapt this for sure to a certain extent to what you want use depending on our vision and how you want to fit it
 
I have absolutely no idea how the technology of DLSS 5 actually works, since Jensen literally spent 30 seconds on it before moving on to the real moneymakers or whatever

It's using Generative AI to add details to each frame in real time and overlaying it over the engine rendered image? How fast is this process? Which Nvidia GPU's have enough TOPS to do this in real time, and at what resolution and frame rate? How is the Generative AI trained, and what is the target endpoint in terms of what details are added? Does the developer have any control over what details are added by DLSS 5? How was the data sourced for this and what is the training used? Does the developer just feed textures and models with 1,000,000x detail into the training dataset and then magic happens?

Jensen's too busy celebrating how much money he made this year to explain anything about this sigh
 
Hugare said:
People not excited about this are stupid or aren't dreaming big enough

This is amazing. Devs will be able to fine tune this, so they could make it work only with environments, or change character models to look good while using this tech etc.

This is like any other use of AI: can be good when used properly.

Two 5090s tho is scary tho. They have to optimize A LOT to make it usable.
Click to expand...

I can see the potential for this. It could be huge having AI transform simple models into realistic ones.
But this demo hits the AI uncanny valley really hard.
 
Nvidia will continue dragging people kicking and screaming into the future whether they like it or not. People used to aggressively despise DLS S and now they absolutely love it, and it's imitators.
 
People will be divisive on if this ruins art direction or not for some games, but man, you can't deny some of these changes enhance the experience for the better. I can't wait to try this when it launches.
 
Last edited:
Looks good for a showing. Its true that, as it is now, it kills some of the artistic intent, but I'm sure there will be customization to it. AC Shadows looked really good, Leon and Grace looked good too and this will only improve.
 
BennyBlanco said:
If you could use this without touching the character models it could be incredible.

This is their first attempt at this. I'm hyped for where it will lead. DLSS1 stunk too and didn't become great until DLSS3 really. As they iterate on it, it will get much better.

I just know from messing around with LLMs on my own PC it is very demanding and requires very good hardware to generate video. I'm very curious if this is running an LLM locally on your hardware to achieve this, or if it's using their data centers. If it's the latter then they're gonna want a sub and I have zero interest in it. If it's the former then you are gonna need 40GB VRAM or something crazy like that. At least as things exist right now.
Click to expand...
Sounds like it's local since they need 2x 5090 for the model to fit so assume it's larger than 32GB just for the model itself.
 
You must log in or register to reply here.

Similar threads

Nvidia Damage Control: Game developers have full, detailed artistic control over DLSS 5 effects
News  2
90
91
viveks86 replied
  • Poll
DLSS 5 - Yes or No?
Opinion Poll  2 3
127
128
viveks86 replied
NVIDIA DLSS 4.5 Dynamic Multi Frame Generation is now due on March 31
Event Platform 
32
227
M.W. replied
Resident Evil Requiem | 4K Path Tracing & NVIDIA DLSS 4 Trailer
Trailer Platform  2 3
115
2K
IbizaPocholo replied
Trouble in Digital Foundry land - Big Johnny boy not happy over DLSS 5 video
2 3
147
148
TheDarkPhantom replied
Top Bottom