DLSS 4.5 Ray Reconstruction Explained

DLSS 4.5 introduced a wealth of new features. A key one of these was the updated Ray Reconstruction. Previously we touched on this but let’s look in more detail at what it is, and what it does for you.

From its name you can take a good guess at what it does, it reconstructs light rays. First introduced with DLSS 3.5 in 2023, its goal was to improve the lighting in game. The problem with ray tracing has often been how many rays of light can be cast in a single frame, this meant that you would get strange artifacts or scenes with a lot of noise. The move was then to use a denoiser.

However, once done the image is upscaled, this could result in artifacts or blurring from the combination of the two processes which wasn't ideal.

The big change here is that while ray tracing still occurs, DLSS 4.5 uses an AI neural network trained using large datasets to help. The AI model takes the ray tracing data and a host of other information such as information from previous frames, where objects are, and where they are moving to, and predicts what is going to happen from this. It uses all that to fill in the gaps and upscale this to match your resolution. The result is a cleaner, more accurately rendered scene without the graininess and unwanted artifacts seen with standard denoise filters and upscaling.

While we have had Ray Reconstruction since 2023 in the form of DLSS 3.5, this second-generation transformer model has up to 35% more compute capacity than the previous version while having no real change in terms of hardware requirement or overhead.

As mentioned, it makes use of previous frames and can predict where an object will be in the next frame. It also uses a bunch of other information from previous frames, the light sources, and in some instances the game data engine itself. As a result, you get much better experience with more realistic lighting, less noise and other oddities that could occur with DLSS 3.5.

This also upscales the image in the within the same process, which is a process is called DLSS Super resolution and, while we had this at the launch of DLSS 4.5, with the addition of ray reconstruction, its full potential is unleashed. To summarize DLSS Super Resolution, its goal is to improve the overall frame rate in your game by rendering the scene in a lower resolution, then upscale it to fit the higher display resolution of your monitor.

The great news here is, Ray Reconstruction works on every NVIDIA GeForce RTX GPU back to the RTX 20-Series. It’s not tied to the newest generation GPUs, meaning that those of you with a 2060 can enjoy significant performance improvements from this at no monetary cost.

To enable this simply launch the Nvidia App, go to settings, then About in the NVIDIA app and hit the check box to opt in to Early Access releases. You will need to ensure you are using at least GeForce Game Ready Driver 580.88 WHQL for this. A full guide on how to enable it per game can be found on the NVIDIA official announcement. Or simply wait for the next driver release in September 2026 when the feature is fully rolled out.

This though does not mean you have all the toys unfortunately, Multi Frame Generation and Dynamic Multi Frame Generation both require the latest RTX 50-series GPUs with Frame Generation requires at least an RTX 40-series card, the chart below breaks this down per Graphics Card series and feature.