DLSS 5 Neural Rendering runs via OptiScaler fork

The development, reported by user ShyVortex, marks a notable shift in how modders are approaching the integration of NVIDIA's upcoming neural rendering features. Previously, the unofficial DLSS 5 implementation relied on a RenoDX-based method that required ReShade to function. The new approach utilizes a fork of OptiScaler, created by developer Dagherbou, which injects NVIDIA's nvngx_dlssnr.dll as an additional processing pass after the game's existing upscaler. This architectural change removes the dependency on ReShade, streamlining the workflow and reducing potential points of failure in the rendering pipeline.

The practical impact of this change is substantial. In testing with The Last of Us Part II, ShyVortex observed frame rates increase from 14 FPS to 30 FPS when switching from the previous method to the OptiScaler fork. This measurement was taken with Neural Rendering enabled in both scenarios and without Frame Generation, effectively more than doubling the performance in a title that is notoriously demanding on hardware. The improvement makes the game playable on configurations that previously struggled to maintain a stable frame rate, highlighting the efficiency gains of removing the ReShade overhead.

The fork also introduces new flexibility in how the Neural Rendering model is applied. It exposes the model resolution as a separate parameter, allowing users to adjust the internal resolution at which the neural network operates. Testing at 4K resolution demonstrated that lowering the model resolution to 75% increased performance while retaining most of the visual quality. Reducing the resolution further to 50% yielded even larger performance gains, though this introduced visible artifacts, particularly around fine details such as character hair. This tunability allows users to balance performance and visual fidelity according to their specific hardware capabilities and preferences.

It is important to note that this is an unofficial implementation and not an official release from NVIDIA. The fork uses the color-composition method developed for the RenoDX DLSS 5 add-on, but the core processing now runs through the OptiScaler framework. Users can swap the DLL files to use versions that already support RTX 40, 30, and 20 series graphics cards. This development occurs ahead of the official NVIDIA release of DLSS 5, raising questions about how the official implementation will compare in terms of performance, compatibility, and feature set. The community's ability to optimize and adapt these technologies before their official launch underscores the ongoing importance of modding in shaping the practical experience of new graphics features.