Andrew Sampson Releases Open-Source DLSS 5 Reimplementation
Developer Andrew Sampson released OpenDLSS-NR, a custom Vulkan reimplementation of NVIDIA's DLSS 5 that enables neural rendering on unsupported hardware.
Andrew Sampson released OpenDLSS-NR, a custom Vulkan reimplementation of the NVIDIA DLSS 5 Neural Rendering network. Hosted on GitHub, the project is a bit-exact reproduction of the 71-block Swin/ViT network from DLSS-NR build 310.8.0. Unlike the official proprietary software, this version operates without NVIDIA libraries and includes a WebGPU port for browser-based execution.
Integrated into the Borderless Gaming utility, the mod allows DLSS 5 features to run on unsupported hardware, including the GeForce RTX 3090. Sampson achieved this by capturing frames at the desktop compositor layer and quantizing the machine learning model to ensure compatibility with uncertified GPUs.
The system also features a Frame Generation capability powered by an independently trained model. Users can adjust rendering parameters, such as artistic style and structural detail, to manage the balance between latency and visual fidelity.
