NVIDIA Introduces 2nd Gen Transformer-based Ray Reconstruction, Now Part of DLSS 4.5

NVIDIA today announced the introduction of 2nd Generation Ray Reconstruction, a component of DLSS 4.5. When it debuted in 2023 as part of the DLSS 3.5 update, Ray Reconstruction addressed a key image quality problem where ray traced shadows, and reflections appeared inaccurate when subjected to DLSS 2 super resolution. In 2025, NVIDIA updated the older convoluted neural networks (CNN) based AI model with a newer transformer based model. Later that year, when NVIDIA introduced the DLSS 4.5 feature-set with 2nd Gen transformer models trained on larger sets of data, Ray Reconstruction didn’t get updated. The company completed the DLSS 4.5 feature-set with new Ray Reconstruction feature. Besides a 2nd Gen transformer model trained on a larger volume of data, DLSS 4.5 RR introduces a more efficient AI denoiser, and greater spatial awareness for lighting accuracy.