DLSS 4.5 Ray Reconstruction is a definite improvement but path traced fidelity is still not a 100% solved issue

Nvidia's DLSS 4.5 Ray Reconstruction feature aims to address the visual artifacting issues commonly found in ray traced or path traced games. While ray tracing delivers highly accurate lighting and shadows, the denoising process used to convert low-fidelity path traced images into fully-realized frames results in visual weirdness during motion.

This is particularly problematic in games with moving objects or characters, leading to visual trails reminiscent of early 20th-century warehouse parties. Ray tracing demands significant computational resources, making it burdensome for most GPUs. The limited number of tracked rays used to approximate scenes leads to a less convincing visual experience.

To tackle this issue, Nvidia introduced Ray Reconstruction, which replaced hand-tuned denoisers with an AI-based denoiser working alongside upscaling models to generate more accurate pixels between sampled rays. The latest version of Ray Reconstruction, employing a sharper second-gen Transformer model, brings four key improvements: an efficient denoiser, enhanced super resolution, expanded training dataset, and finer developer control.

These enhancements make Ray Reconstruction more effective in producing stable images with improved lighting accuracy and temporal stability, while retaining more detail in the scenes.