Nvidia has detailed the many new features and improvements of DLSS 4, which promises increased performance, but also better rendering quality, and not only for the next GeForce RTX 50. Here is what this big update offers.
Alongside the long-awaited announcements of its Blackwell architecture graphics cards, Nvidia has finally said more about DLSS 4, the new major version of its upscaling technology. Nvidia announces it, this is the biggest evolution of DLSS since 2020 and its version 2.0.
Very little information has leaked about this update, which promises to improve almost all of its components, between image generation, ray reconstruction and even the general image quality of the tool. Here’s everything DLSS 4 brings to Nvidia GPUs.
Image generation taken to the next level
The key new feature of DLSS 4 is multi-image generation, a feature that adds not one but up to three additional images generated using AI.
Available only on GeForce RTX 50 graphics cards, this evolution of image generation promises almost doubled performance compared to the old version. A new feature which is made possible in particular by the new Tensor cores which offer more than 2.5 times AI computing power compared to the previous architecture.
We can then wonder about the impact of such a feature on the overall latency of the game. Nvidia claims to have worked first of all on the overall fluidity of image generation, with better managed frame-pacing, but also on Reflex 2, the new version of its latency reduction technology.
In addition, Nvidia has improved its classic image generation model (for the RTX 50 and 40) to make it faster, but also more efficient. We can therefore expect better performance (up to 40%) with a memory footprint reduced by 30%.
Improved image quality for all RTX
As Nvidia points out, DLSS is a technology that by nature continually improves, as long as its model continues to be trained.
With DLSS 4, Nvidia has improved the rendering quality of Super Resolution, ray reconstruction and even DLAA. This improvement was made possible thanks to the transition to a so-called transformer model, offering double the parameter compared to a CNN (Convolutional Neural Network), used until now by Nvidia, but also by Sony on the PS5 Pro.
This new model promises better image stability when DLSS is activated, thus reducing possible ghosting effects in motion, but also flickering phenomena in the most complex scenes.
Ray reconstruction, which already improves the rendering of ray tracing in compatible games, also benefits from these rendering improvements. This is also the case for DLAA, Nvidia’s AI-boosted anti-aliasing, which should also improve stability in motion.
A DLSS 4 in beta version for the moment
This new model will however not be activated by default and will first be available in beta. Gamers will be able to activate it via the Nvidia App to test the improvements before it is rolled out more widely for all GeForce RTX graphics cards.
75 games will be compatible with multi-image generation when the GeForce RTX 50 is launched, including Indiana Jones and the Elder Circle, Cyberpunk 2077 and Star Wars Outlaws. In the future, competitive games like Marvel Rivals and Nakara: Bladepoint may benefit from this.
Find all the new features from CES 2025