An independent developer has managed to get Nvidia DLSS 5 running on AMD Radeon hardware. The community project adapts leaked neural network model files specifically for RDNA 4 cards like the Radeon RX 9070 XT. Early benchmarks show heavy performance penalties, but recent code updates are already clawing back lost frames.
Created by a developer going by danielblnc, the project is called DLSS NR on AMD. Rather than swapping out the algorithm with FSR or another upscaler, the tool directly translates the underlying neural network weights for AMD silicon. Testers confirmed the software generates a custom file named dlssnr on amd weights bin. This proves the card is actually executing the neural model rather than faking compatibility through a wrapper.
Getting the code to run is 1 thing, but getting playable frame rates is a different story. In Cyberpunk 2077 at 1440p resolution, frame rates tanked from 86 fps down to 12 fps during initial tests. Things are improving quickly though. The newer version 0.2.7 release pushes performance up to around 30 fps at 1080p, representing a 12% speed boost over the previous build.
The mod still suffers from major architectural bottlenecks on Radeon silicon. The developer plans to target register pressure and local memory constraints in future releases to free up compute resources. Better compute unit usage and native FP8 support will also be essential if this experimental port ever hopes to match native upscaling speeds.
