It was recently reported that developer MAAN has released OpenDLSS-NR, an open-source Vulkan reimplementation of NVIDIA's DLSS 5 neural rendering network.
The core breakthrough of this project is that it enables DLSS 5 to run on GPUs other than the RTX 50 series for the first time.
OpenDLSS-NR precisely replicates the 71-block neural network structure found in DLSS 5 runtime DLL version 310.8.0. The developer claims its output is "bit-identical" to NVIDIA's official implementation. Not only is the final image identical, but the outputs at the boundaries of all 75 intermediate blocks match byte-for-byte.
The network employs a U-Net architecture, combining shifted window Transformers with global vision Transformers. It uses FP8 activation values and FP16 accumulation precision, with model weights totaling approximately 141 MiB.
In terms of performance, data measured by MAAN using an RTX 4070 SUPER shows execution times of just 2.8 ms at 768×768 resolution, 7.8 ms at 1080p, 12.6 ms at 1440p, and 29.3 ms at 4K. These figures indicate that the full DLSS 5 network can be executed efficiently on the Ada Lovelace architecture.
Even more groundbreaking is the WebGPU port. This version runs entirely in the browser without relying on NVIDIA Tensor Cores or native FP8 support. At 512×512 resolution, the execution time is 72 milliseconds. While significantly slower than the 2.7 milliseconds of the native implementation, it sufficiently demonstrates that DLSS 5's neural network can be decoupled and run on non-NVIDIA hardware.
Currently, the project has limitations. The repository does not include NVIDIA's proprietary model weights, so users must provide their own. It is not a plug-and-play game mod, but rather a standalone implementation accompanied by a demo renderer based on the Google Filament engine. It supports motion vectors and temporal feedback but does not implement DLSS super-resolution functionality.
For RTX 40 series users unable to use official DLSS 5, this project presents another possibility. Although it is still some distance from actual game deployment, it proves that the core technology of DLSS 5 does not rely entirely on the latest hardware.