- Yes
- No
Pro’s:
- Improved render performance by replacing taxing multi-layer denoisers with one single neural denoiser.
- Improved fidelity of reflections and shadows by accounting for lighting effects in the denoising process.
- Evolving capability - Ray reconstruction is a key function of DLSS, and is consistently updated with new improvements and presets that can be updated by the developers or playerbase through DLSS overide.
Con’s:
- Requires dev time for implementation In-Game
- Sometimes produces artifacts
- Sometimes smooths textures unnecessarily
The use of Nvidia’s own denoising tech could improve performance as seen in cyberpunk, or offer improved fidelity, or even both.
Standard denoisers can often remove certain lighting effects from a scene like ambient occlusion because it’s trying to “smooth out” a lighting scene, whereas the DLSS model is trained to identify global illumination features and keep them in the scene, whilst removing visual noise and reducing ghosting.
The most noticeable improvement will be with RT Ambient Occlusion Shimmering, which on “performance” mode is dire, and in quality mode still leaves some shimmering at an fps drop.
Comparison of current denoisers:
As shown in the two video’s, both denoisers in game produce a level of shimmering and artifacting, however the quality preset massively reduces visual noise. However, Ray Reconstruction could offer further reduced visual noise to improve the games graphical performance.
Performance:
Quality:
How AI ray reconstruction works:
Spoiler
NVIDIA DLSS 3.5 | New Ray Reconstruction Enhances Ray Tracing with AI
An example of how DLSS Ray Reconstruction can improve fidelity:
