Community developers have found a way to run NVIDIA’s DLSS 5 Neural Rendering on some AMD Radeon GPUs using custom HIP-kernel runtimes and game-integration tools. It is an unofficial community effort—not NVIDIA support for Radeon—and the reported performance gains are early results, not a promise of higher frame rates in every game.
2
4
14
How the Radeon implementation works
NVIDIA describes DLSS 5 as a rendering stage for GeForce RTX 50 Series GPUs.
14 The AMD projects take a different route: they reimplement the neural-network runtime for Radeon hardware using HIP kernels, then connect it to supported games through their rendering and upscaling paths. The project documentation describes support for RX 9000 and RX 7000 cards in DirectX 12 games that use FSR.
2
4
That distinction matters: this is not an official NVIDIA port or a general compatibility layer. It is a community-built runtime and integration that can work only where the GPU, game and setup meet its requirements.
2
4
Daniel Blanco’s closed runtime vs. Kien’s open-source option
The two runtimes differ most clearly in transparency. Daniel Blanco’s DLSS-NR-on-AMD runtime is distributed as a closed-source binary, while Kien’s alternative is described as open source. Both are associated with HIP-based AMD support; the available reporting does not establish that either runtime is consistently faster or produces better image quality.
6
7
35
AMDNR Launcher supports both runtimes, allowing users to select one through the launcher rather than treating them as a single implementation.
35 The distinction is useful for people who want to inspect or modify open-source code, but the sources provided do not offer a controlled, head-to-head comparison of the two options.
What the reported performance numbers show
The developer’s successive releases reported a 23% performance improvement in Alpha 0.3.3 over 0.3.2, then a further claimed 42% in Alpha 0.4.0 over 0.3.3. Release notes also reported an 8% improvement for 0.4.1 over 0.4.0.
3
6
15 These are version-to-version claims, not independent tests of the two runtimes, and they should not be combined into a universal expected gain.
Separate early reports placed Cyberpunk 2077 on an RX 9070 XT at roughly 30–34 frames per second at 1080p, but the figures come from different builds and test conditions.
1
4 They provide a snapshot of development, not a like-for-like benchmark or a reliable prediction for another game, settings profile or GPU.
Radeon and handheld support
The project lists Radeon RX 9000 (RDNA 4) and RX 7000 (RDNA 3) support, subject to the game and FSR requirements.
2
4 A separate report describes an experimental one-click handheld profile benchmarked on an ASUS ROG Ally with a Z1 Extreme chip.
32 That is evidence of an early test on one device configuration, not proof of broad handheld compatibility or consistent performance across games.
What AMDNR Launcher and OptiScaler change
Installing an injection-based mod can involve setting up game files and choosing compatible components. AMDNR Launcher automates much of that process, with reporting describing game and runtime selection, component installation, removal and runtime updates through its interface.
6
35
OptiScaler is part of the integration picture: a neural-rendering fork adds support for the feature in DirectX 12 games, while the launcher provides a more accessible way to deploy and select runtimes.
18
35 The launcher can reduce setup friction, but it cannot make an unsupported game compatible; game support and the required rendering path still matter.
4
18
Limitations to keep in mind
Performance remains a work in progress, and the project documentation says some effects may not work fully.
4 Reports also warn that injection can be blocked by anti-cheat systems, so the safer use case is single-player experimentation—not protected multiplayer games.
6
10
The most reliable takeaway is that community modders have made Radeon support possible, while the experience is still experimental. Runtime choice, game compatibility, visual results and frame rates need to be checked individually; the available evidence does not show that either runtime is a settled winner.