OpenDLSS: A Vulkan Reimplementation of Nvidia's DLSS 5 Neural Rendering Network
The repo claims byte-for-byte parity with Nvidia’s DLSS 5 neural rendering network, including intermediate block boundaries.
OpenDLSS-NR implements the 71-block Swin/ViT DLSS-NR network in Vulkan, using FP8 on tensor cores and requiring users to supply their own weights. It says the model is neural rendering rather than upscaling: input and output stay at the same resolution while the network re-renders detail, tone, structure, and skin from frame inputs and noise. A separate WebGPU port is included and is described as matching the same bytes in-browser without tensor cores or FP8. The project says it is unaffiliated with Nvidia and includes no Nvidia software, weights, or instructions for obtaining them. HN · GitHub AI's note
OpenDLSS-NR implements the 71-block Swin/ViT DLSS-NR network in Vulkan, using FP8 on tensor cores and requiring users to supply their own weights. It says the model is neural rendering rather than upscaling: input and output stay at the same resolution while the network re-renders detail, tone, structure, and skin from frame inputs and noise. A separate WebGPU port is included and is described as matching the same bytes in-browser without tensor cores or FP8. The project says it is unaffiliated with Nvidia and includes no Nvidia software, weights, or instructions for obtaining them. HN · GitHub AI's note
score 5
Discussions
- hn · 273 points · 125 comments