NVIDIA's Deep Learning Super Sampling (DLSS) technology has long been the gold standard for AI-driven upscaling, but it has traditionally been exclusive to GeForce RTX graphics cards. However, a significant development in the open-source community is challenging this boundary. A new release of the d4r project has introduced experimental support for NVIDIA DLSS 4 and DLSS 4.5 on AMD Radeon RX 9000 series GPUs. This breakthrough allows AMD hardware to utilize NVIDIA's latest neural upscaling models, marking a notable shift in the competitive landscape of PC graphics technology.
How d4r Bridges the Gap Between AMD and NVIDIA
The d4r project, developed by a creator known as "countervolts," functions by running NVIDIA's official nvngx_dlss.dll through a specialized software stack designed for Linux and Proton environments. The core mechanism involves using ZLUDA to translate CUDA workloads, which are native to NVIDIA GPUs, into instructions that AMD hardware can execute. Additionally, the project incorporates custom kernels for RDNA 3 and RDNA 4 architectures to optimize specific parts of the DLSS pipeline.
Previously, running NVIDIA's upscaling on AMD cards required converting FP8 (8-bit floating point) calculations to FP16 (16-bit floating point), a process that incurred a significant performance penalty. However, the latest update leverages a key architectural advantage of the RDNA 4 generation. The Radeon RX 9000 series GPUs, including the RX 9060 and RX 9070 models, possess native support for FP8 calculations. This means the d4r project no longer needs to perform costly data type conversions on these specific cards, potentially reducing the overhead associated with running DLSS on AMD hardware.
Experimental Status and Current Limitations
Despite the technical promise, it is crucial to understand that this implementation remains strictly experimental. The release notes for d4r versions 0.1.2 and 0.1.3 explicitly state that the runtime and performance on RDNA 4 hardware are currently unverified. While the software lists the Radeon RX 9070 XT, RX 9070, and other RDNA 4 GPUs as supported targets, the developer has only completed testing on the previous-generation Radeon RX 7700 XT.
RDNA 4 runtime and performance remain unverified on the hardware, and only the Radeon RX 7700 XT has been tested by this project.
This distinction is vital for potential users. The project serves as a proof of concept rather than a stable, ready-to-use solution. Enthusiasts who attempt to use this mod should expect potential instability, compatibility issues, or suboptimal performance until further testing is conducted on the actual RX 9000 series hardware.
Performance Implications for RDNA 4
The ability to execute FP8 natively is a significant technical milestone. In earlier iterations of such cross-compatibility projects, the conversion of data types often negated the benefits of using a superior upscaling model. By bypassing this conversion step on RDNA 4 cards, d4r aims to deliver a more efficient execution of DLSS 4 and 4.5. The developer has already demonstrated that native kernels on the RX 7700 XT can improve performance compared to running purely translated code, suggesting that the RX 9000 series could see similar or even greater gains due to its architectural improvements.
Practical Impact for PC Builders
For the average PC builder, this development does not immediately change the recommendation to use AMD's native FidelityFX Super Resolution (FSR) for daily gaming. FSR remains the supported, stable, and optimized solution for Radeon GPUs. However, the existence of d4r highlights the increasing parity between AMD and NVIDIA in terms of AI capabilities. It suggests that the hardware gap is narrowing, and that AMD's newer architectures are becoming capable of handling complex AI workloads that were previously the exclusive domain of NVIDIA's Tensor Cores.
This project also underscores the power of the open-source community in the PC hardware space. By reverse-engineering and adapting proprietary software, developers are enabling features that manufacturers may not officially support. This could pressure both AMD and NVIDIA to further enhance their native AI upscaling technologies to maintain a competitive edge.
What to Watch Next
As the d4r project evolves, several factors will determine its long-term relevance. First, the community will need to verify the performance claims on actual RX 9000 hardware. If the native FP8 support delivers a tangible frame rate boost or quality improvement over FSR, it could become a niche but valuable tool for enthusiasts seeking the best possible image quality in specific titles.
Second, we should watch for AMD's response to these developments. While AMD has not officially commented on this specific mod, the ability to run NVIDIA's DLSS on their hardware may influence their own roadmap for AI features. Finally, the stability of the Proton and ZLUDA stack will be critical. If the project matures into a more reliable solution, it could offer users a hybrid approach, allowing them to choose between AMD's FSR and NVIDIA's DLSS depending on the game's specific requirements.
In conclusion, the arrival of DLSS 4.5 on Radeon RX 9000 GPUs via the d4r project is a fascinating technical achievement. While it is not yet a replacement for FSR, it demonstrates the growing flexibility of modern GPU architectures and the innovative potential of the enthusiast community. For now, it remains a compelling proof of concept that warrants close attention from technical readers and PC gamers alike.