Open-source project claims byte-exact Vulkan implementation of NVIDIA’s DLSS 5 network
OpenDLSS-NR also describes a separate browser-based WebGPU version, but users must provide model weights and the project does not implement DLSS Super Resolution.
OpenDLSS-NR, a GitHub project listed by Hacker News, describes a Vulkan implementation of NVIDIA’s DLSS 5 neural rendering network alongside an independent browser-based WebGPU implementation.
The project claims its Vulkan version reproduces the network’s outputs byte for byte at all 75 block boundaries. The claim is not independently verified in the supplied material; the repository says its reference captures are not included.
OpenDLSS-NR describes the network as 71 blocks and says it re-renders an engine’s frame at the same resolution. It is therefore not an upscaler. The project says its demo includes a temporal feedback path, while a separate command-line tool runs single frames without history.
Users must supply the model weights in a specified directory layout. The repository says it contains no NVIDIA software or weights and is not affiliated with, endorsed by or supported by NVIDIA.
DLSS Super Resolution, a different network, is not implemented. The project also says it refuses models with a different block count from the one it describes.
