Nvidia gave a first glimpse of DLSS 5 this week at its GTC conference. Here’s everything we know so far about this highly discussed technology and answers to the questions everyone has.
Nvidia presented DLSS 5, the next version of its flagship technology for RTX GPUs. The announcement sparked unprecedented controversy for Nvidia, ranging from mockery of the examples given to confusion about how it works.
Many misconceptions are circulating about DLSS 5, and we wanted to address pressing questions about this cutting-edge technology. We are basing our information solely on Nvidia’s statements and engineers’ reports from the GTC 2026. Please note that we are making assumptions, especially when information is officially unknown.
At present, there are many unknowns about DLSS 5 as the technology is not expected for many months. Nvidia assures us that the upcoming weeks and months will focus on explaining the technical challenges, with a scientific white paper logically expected to be published as its public deployment approaches.
What is DLSS 5?
DLSS 5 is the next major update to DLSS, Nvidia’s upscaling and image generation technology for GeForce RTX graphics cards. This new version adds a new component: a real-time neural rendering model, which currently does not have a name.
This functionality of DLSS 5 is based on generative AI to inject realistic lighting and materials into video games. Nvidia’s goal is to achieve visual fidelity previously reserved for Hollywood special effects.
How does DLSS 5 work?
DLSS 5 works in addition to the traditional rendering of shader cores, thanks to the AI accelerators of RTX graphics cards, the Tensor cores. To apply its neural rendering, the technology takes into account color data from an image (the color buffer) as well as motion vectors of a scene.
In simpler terms, DLSS is based on a 2D image of a scene using its color data and direction of its pixels compared to the previous image. Based on this data on a 2D plane, an AI model analyzes the scene’s semantics to infer its properties: it can detect metallic surfaces, organic matter, a character’s skin, a tree branch, or fabric.
DLSS 5 is particularly based on scene lighting to generate a more realistic rendering and precise material reactions. It can add different light effects and other treatments, such as subsurface scattering to simulate light penetration under the skin or translucent objects.
DLSS 5 does not access this data via the game engine but through its own image recognition model trained on millions of other visuals.
DLSS 5 does not alter the geometry of the 3D scene. The technology applies this rendering in an overlapping manner, remaining firmly rooted and faithful to the original 3D geometric models provided by the game engine.
Is DLSS 5 just a generative AI filter?
Unlike a simple filter (like on social networks, but also with tools like Reshade) that would apply a visually incorrect or random layer, the unified model of DLSS 5 is deterministic and promises temporal stability from one image to another thanks to motion vectors.
It truly “understands” the semantic of a scene rather than applying a simple filter that would resemble an overlay.
DLSS 5 does not perform rendering without conditioning, that is, without the information provided by the traditional rendering of a 3D scene. According to Nvidia, the rendering quality will depend on the quality of the input data: a game with complex 3D models with path tracing light rendering will theoretically have a better result once DLSS 5 is activated.
Will DLSS 5 be compatible with all DLSS-supported games?
Like the current DLSS, DLSS 5 will be implemented on a case-by-case basis by game studios and developers, in collaboration with Nvidia.
Like Frame Generation or Ray Reconstruction, this component of DLSS 5 will require extra work and will not be automatically compatible with games already supporting DLSS.
Note that DLSS 5 could be used by modders to implement the technology in games that do not natively support it.
How will developers and artists be able to adjust DLSS 5?
Developers will have specific controls including intensity adjustment, color grading, and masking. Masking is particularly important: it allows creators to define which parts of the image (e.g., a specific character’s face) should not be affected by DLSS 5, theoretically ensuring the respect of their artistic vision.
We currently do not know how these different settings will materialize for developers. One unknown is whether they can adjust the intensity and color grading of DLSS 5, can they adjust its fundamental nature? Specifically, can developers define how the technology should improve the scene rather than simply adjust the intensity? Many questions that should have answers in the coming months.
Can DLSS 5 be disabled?
Yes, DLSS 5 will be a completely optional feature that players can choose to enable or disable in the game settings, like the rest of the DLSS options.
On which graphics cards will DLSS 5 work?
DLSS 5 is extremely resource-intensive and will work on high-end GeForce RTX 50 series graphics cards and above.
During its presentation, the technical demo actually required the simultaneous use of two RTX 5090 to run in real-time.
However, Nvidia has clarified that the final version will work on a single graphics card.
When will DLSS 5 be released?
The launch of DLSS 5 is scheduled for fall 2026. Note that Nvidia has only presented one facet of DLSS 5 in its presentation, and other features of this new update should be presented in the coming months.
