DLSS 4.5 Ray Reconstruction Arrives in August with a New Second-Generation AI Model

NVIDIA has announced the arrival of DLSS 4.5 Ray Reconstruction, a new evolution of its neural rendering technology that will be available throughout August. The update is part of the broader DLSS 4.5 ecosystem, which has introduced innovations over the past few months, such as the second-generation Transformer model for Super Resolution, Dynamic Multi Frame Generation, and further improvements to frame generation techniques.

Ray Reconstruction represents one of the most advanced elements of the DLSS suite. The technology is designed to enhance visual quality in scenes processed through ray tracing and path tracing, replacing traditional manually developed denoising algorithms with an AI model trained on NVIDIA supercomputers. The goal is to more accurately reconstruct missing information in areas of the image where the number of sampled rays is insufficient to achieve a clean result.

According to NVIDIA, the new second-generation Transformer model introduces significant improvements in various aspects. The company claims that the system has 35% more computational capacity and manages 20% more parameters compared to the previous generation while maintaining similar performance in terms of execution impact.

Among the most relevant new features is greater spatial awareness of the scene and a more sophisticated use of data from the graphics engine, including pixel sampling and motion vectors. This should lead to a more accurate image reconstruction, with more faithful lighting, better temporal stability, and a cleaner representation of movement in content that makes extensive use of ray tracing and path tracing.

NVIDIA also highlights that the new model has been trained on a larger dataset, a characteristic that should enable the algorithm to more accurately identify useful information for image reconstruction and bring the final result closer to the so-called "ground truth," which is the quality reference obtained through extremely accurate rendering.

The update is not limited to the gaming sector. The company has confirmed that DLSS 4.5 Ray Reconstruction will also be integrated into Blender Cycles as a new denoising system. In this context, the technology aims to enhance the experience of 3D artists while working in the viewport, allowing them to view images close to the final rendering quality without compromising the interactivity of the scene.

Integration into Blender represents the evolution of the previous OptiX-based support and should enable faster workflows in the development phases of lighting and visual appearance of scenes. NVIDIA expects this feature to be rolled out alongside Blender 5.3, which is anticipated to be released in the autumn.