NVIDIA has officially announced that DLSS 5 will make its debut tomorrow with the release of NBA 2K27, whereby AI can potentially improve scene rendering in games. The DLL files for DLSS 5 recently leaked online after being discovered in an early build of the next NBA title from 2K and Visual Concepts. Modders were quick to begin posting video comparisons showing it in use with dozens of games, and most, but not all, showed promise in its ability to add photorealistic upgrades to games. NVIDIA originally revealed DLSS 5 to the world back in March, and one of its first criticisms was that it could potentially make drastic changes to a game, which detracted from an artist’s or developer’s vision. NVIDIA added that, ultimately, developers could make the choice of how DLSS 5 was implemented and said they would be presented with a range of settings rather than a static configuration that would be fully controlled by the graphics card.
“What’s great is the level of control that it gives us. We can set the overall tone and style to match our existing art direction and then really get precise where it matters most, using a per pixel uplift control mask to fine tune detail on characters. Given that we want to respect their likeness with the utmost care.”
Peter Kavic, Senior Producer – Visual Concepts
Per the gameplay footage seen in the above video, to say that DLSS 5 brings photorealistic image quality to the game could be a major understatement. If what we see in NBA 2K27 is the beginning of the DLSS 5 train, it will be exciting to see what comes next, and needless to say, 3D-Guided Neural Rendering for the masses has officially arrived.
While modders have managed to get DLSS 5 working on RTX 30 and 40 series GPUs, usually with major performance impacts, NVIDIA has said that it is launching for the RTX 50 series desktop and laptop GPUs and will also be available to GeForce NOW Ultimate subscribers.
“On September 3rd we’re releasing DLSS 5, which introduces 3D-Guided Neural Rendering, a major step forward in using AI to bring real-time graphics closer than ever to Hollywood-grade photorealism.”
– NVIDIA
Discussion (19 replies)
Join Discussion →"to say that DLSS 5 brings photorealistic image quality to the game could be a major understatement"
Surely you mean overstatement, right?
I don't know, the footage in that video was a wow for me. I didn't expect it to look that good but I didn't watch enough to confirm if a pre-rendered cutscene or render in game.
Honestly this looked like LA Noire to me, straight from the uncanny valley. Some of the other games modders unofficially enabled it on look far better to me.
I like how they are trying to highlight how much control the developer has over the process, so it doesn't just sound like it's AI slop.
Here's what I don't understand. Presumably the AI isn't creating something from nothing. Enhanced views of veins and scarring and such being more prominent has to be data that is there that is being made more highlighted... right? If that is the case WTF are the dev's not doing that from the jump?!
That's a good point... why not run your textures through DLSS 5, then just revise your textures to match the AI-enhanced content
Those extra clicks would take time away from minecraft/fortnite so not worth the extra effort?
This has been my main argument. The developers can already create high quality characters. Why not just do this on the developer side, sans DLSS5/AI.
Just natively create higher quality NPCs in the game /shrug
This is why quality settings exist in the game anyway. Create a new graphics setting in-game called "Photorealistic" or something fancy, that sits above "Ultra" settings, for people that want to cut their framerate in half, but want the absolute best quality the developer can provide. Then leave the settings under that, the more LOD'ed out versions, less detail, that run fine.
I dunno, I just don't see the point, why put the burden on NVIDIA/AMD to use AI to up the detail of games, and not the developer who is making the game themselves?
This is a filter that is applied after the regular rendering passes, meaning you can't do it prior to rendering the scene. It is applied to a whole scene not just assets.
Oh, but it does, create something from nothing. It detects a human face it tries to hallucinate a photoreal version of it. Sometimes it works out great, sometimes it doesn't.
I guess my point is, developers can already just add this much detail to their characters in the first place.
They don't really need DLSS 5 to make their characters look better. They can just create them that way in the game design to begin with.
Remember when DLSS meant improving performance? Peppridge Farm remembers.
Totally impressive id say
I wonder how much additional overhead this is to be 'generating' the more detailed or 'realistic' look for characters? Personally... I'm not looking to use DLSS5 to make my games slower in order to alter the vision of the game..
BUT... if properly integrated into the spaces that make sense in the game by the developers and tuned... I bet the implementation could be much more refined.
[embedded media]
I mean it IS neat that it can do this even at 50% performance but I wonder about the power impact.
[URL unfurl="true"]https://www.techpowerup.com/352262/early-dlss-5-testing-suggests-the-rtx-5090s-single-power-connector-might-be-the-bottleneck[/URL]
All you peasants with 600W max 5090 cards can suck it! :D
Time to invest in specially modded 1000W 5090.