Version 0.4.3 is a binary alpha, not an open-source project
DLSS-NR on AMD makes an arresting promise: run NVIDIA's DLSS 5 Neural Rendering effect on an AMD Radeon card inside a game that exposes FidelityFX Super Resolution inputs. The author describes it as a fresh implementation of the neural-network runtime, including HIP kernels, memory layouts, and game integration. The GitHub repository does not let you inspect any of that work. Its tree contains a README, license, third-party notice, and funding file. The working artifact arrives only as a Windows executable attached to each release.
That matters because the license reserves the source, forbids modification and reverse engineering, bans redistribution, and limits use to personal, non-commercial machines. Version 0.4.3's installer is 15,910,400 bytes, while the repository has 0 implementation source files. Calling this open source would be inaccurate. It is a free-to-use proprietary alpha distributed through GitHub, with Microsoft Detours and AMD FidelityFX headers identified separately under their own licenses.
The supported path is Windows 11, DirectX 12, and recent Radeon hardware
The installation instructions are short. Put dlssnr_on_amd_setup.exe in the game's executable directory, add your own legitimate nvngx_dlssnr.dll build 310.8.0.0, run the setup program, launch the game, and enable FSR. The End key opens an overlay for mode and image controls. Running the installer again provides update and removal actions. No hosted account or API token is part of the documented path.
Compatibility is the difficult part. The README requires Windows 11, a DirectX 12 game with FSR, Adrenalin 26.1.1 or newer, and an RX 9000 or 7000 GPU. Only an RX 9070 XT is named as tested hardware. The author asks RX 7000 users to submit reports, says older cards are unsupported, and describes Vulkan as a hope for a later release. Cyberpunk 2077 and GTA V Enhanced are the two named test games. Anti-cheat blocks the DLL.
What happened when we ran it
Our lab has no install, build, or test result for commit 057c873. The automated runner found no supported ecosystem because GitHub reports the primary language as null, and the repository supplies no Dockerfile. The sandbox is an unprivileged container without the required Windows 11 host, DirectX 12 game, Radeon GPU, Adrenalin driver, or user-supplied NVIDIA DLL. No package count, audit result, or timing exists to report.
That absence limits the verdict. We did not measure frame rate, image quality, input latency, game compatibility, installer behavior, or malware status. The README currently says performance is a work in progress and gives an RX 9070 XT figure, while the v0.4.3 release notes claim a much higher result for one Cyberpunk 2077 mode. Both numbers belong to the author. Neither is a result from our sandbox, so buyers should reproduce the exact game, resolution, mode, driver, and GPU combination themselves.
Rapid releases coexist with 165 open issues
The project was created on September 3, 2026 and published 23 alpha releases by September 28. The latest release arrived the same day as the last repository push. That is active development, and the release notes show repeated work on performance, overlays, and settings. GitHub also counted 167 open issues and pull requests when checked: search separated them into 165 issues and 2 pull requests. A fast release cadence has not yet produced a settled compatibility table.
Current reports describe the kind of risk an injected graphics alpha carries. Issue 206 provides logs for a crash when staging is recreated while the overlay is open and includes a workaround involving configuration edits with the overlay closed. Other open reports cover the End-key overlay failing to appear, specific games refusing to start, and visual artifacts. These are user reports, not findings from our lab, but they give competent testers concrete failure modes to expect.
Use it as a reversible experiment on one supported game
The best candidate is an RX 9000 owner with a backed-up game folder, anti-cheat disabled, the required DLSS DLL obtained legally, and patience to inspect dlssnr_on_amd.log. Start with one of the 2 games named by the author, record the driver and mod version, and compare screenshots or captures rather than trusting a first impression. Keep the installer available so removal is immediate if the game crashes or updates break the hook.
Everyone else has a cleaner option. OptiScaler exposes source and targets broader upscaler replacement. DLSS Swapper manages vendor DLL versions on compatible systems. Game developers can integrate the FidelityFX SDK directly. DLSS-NR on AMD is interesting precisely because it crosses a vendor boundary, but v0.4.3 asks the user to trust an unauditable executable inside a game process. That trade is reasonable for an experiment. It is a poor fit for a stable gaming setup or any development workflow that requires source review.