Nvidia has been releasing major improvements to its DLSS suite of neural rendering tech throughout 2026. The first major update came at CES with DLSS 4.5 upscaling (or "Super Resolution"), which introduced a more advanced and more computationally intensive transformer AI architecture for better image quality at lower input resolutions. The second was DLSS 4.5 Multi Frame Generation, which introduced 5x and 6x multipliers along with a dynamic mode that shifts multipliers on the fly for the smoothest gameplay experience.
Now, at Computex 2026, Nvidia has revealed DLSS 4.5 Ray Reconstruction, its advanced denoiser for better ray-traced and path-traced image quality in games.
For a refresher, DLSS Ray Reconstruction replaces hand-tuned denoisers with a neural rendering model that both infers what pixels should look like in noisy areas where light rays weren't cast and upscales the resulting image to a higher-resolution output. The model uses temporal and spatial inputs from game engines to produce its sharp, stable, and high-fidelity output images.
DLSS 4.5 Ray Reconstruction, coming in August, improves this set of techniques in three ways. Using an improved transformer architecture, it can process 35% more input data and uses 20% more parameters within the same compute budget as the previous-generation transformer architecture.
DLSS 4.5 Ray Reconstruction also inherits better spatial awareness and processing of input data from DLSS 4.5 Super Resolution, resulting in more accurate lighting, more stable images, and clearer motion. Nvidia says the mew model has also been trained on a larger data set to improve its utilization of game inputs and offers developers more control over its temporal accumulation behavior, both of which result in improved image quality.
Unlike DLSS 4.5 Super Resolution, which incurs substantial performance penalties on older hardware, DLSS 4.5 Ray Reconstruction will remain 100% compatible with every GeForce RTX GPU, including RTX 20-series and RTX 30-series products.
Nvidia touts a couple of scenarios where DLSS 4.5 Ray Reconstruction makes a big difference in image quality, and the improved performance of the model is indeed easily visible.
In Pragmata, the flickering of a laser trap becomes much more dynamic, and it doesn't leave persistent artifacts when those same lasers deactivate.
And in Alan Wake II, a wall of TVs displaying nothing but static transforms from a mere suggestion into a crisp, convincing reproduction of CRT snow. And the reflections from those TVs on the nearby floor also becomes much more dynamic and lifelike.
Ray Reconstruction has always been the most specialized of the DLSS model family, so it's perhaps unsurprising that it's the last of the three to be updated this year. Nvidia says it'll come to 27 games in August via the Nvidia App, including hits like Cyberpunk 2077, Hogwarts Legacy, Pragmata, and Resident Evil Requiem. And of course, it's fully compatible with the Blackwell GPU on the RTX Spark.
The benefits of Ray Reconstruction will soon move beyond games, too. Nvidia is bringing this model to Blender, where it will replace slow denoisers that incur significant wait times for the viewport to re-stabilize every time the camera is moved. Along with Nvidia Optix, Ray Reconstruction in Blender promises a much more interactive viewport experience and near-final image quality, resulting in a better representation of one's creative vision and less disruption to the creative flow state.
We'll be going hands-on with some of Nvidia's latest products and releases at Computex 2026, and it's likely we'll be able to see DLSS 4.5 Ray Reconstruction in action. Stay tuned for more details as we get them.
