Early Access to DLSS 5 on AMD Hardware
A dedicated modder has successfully integrated Nvidia's Deep Learning Super Sampling (DLSS) 5 technology into a form that can run on AMD's RDNA 4 GPUs. This achievement, while in its nascent stages, represents a significant, albeit unofficial, step towards cross-vendor AI upscaling capabilities. The initial implementation targets the upcoming RX 9070 XT, with the goal of eventually reaching performance parity with Nvidia's own DLSS 5 implementation on comparable hardware, such as the RTX 4050 Ti.
The modder, known within the community for pushing hardware boundaries, has made the necessary files available for adventurous users. However, the current state of the mod is far from optimized. Early testing on the RX 9070 XT yields a mere 30 frames per second at 1080p resolution in supported titles. This performance level is considerably lower than what would be considered a smooth gaming experience, and significantly undercuts the performance expected from a GPU of the RX 9070 XT's purported tier. The excerpt from Tom's Hardware wryly notes that users might "absolutely destroy the stable performance you were getting before," highlighting the current instability and performance degradation.
The Technical Hurdles of Cross-Vendor AI Upscaling
Nvidia's DLSS technology relies on specialized Tensor Cores within its RTX GPUs for AI inference. These cores are designed to accelerate the matrix multiplications fundamental to neural network operations. AMD's RDNA 4 architecture, while featuring significant advancements in its own right, does not possess direct equivalents to Nvidia's Tensor Cores. This fundamental hardware difference means that running DLSS 5 on AMD hardware requires a workaround, likely involving the use of general-purpose compute units (like stream processors) or potentially dedicated AI accelerators if RDNA 4 includes any such nascent hardware. This translation layer inevitably introduces overhead, explaining the current performance penalty.
The modder's success suggests that the core DLSS algorithms, when abstracted sufficiently, can be executed on alternative hardware. However, the efficiency of this execution is currently very low. The process likely involves complex emulation or re-routing of computation, which taxes the RDNA 4 GPU's resources heavily. Achieving the target performance of a 5070 Ti (which would imply a significant leap from the current 30 FPS) will necessitate substantial optimization, potentially requiring deeper insights into both DLSS 5's internal workings and RDNA 4's specific compute capabilities. It's akin to trying to run a highly specialized piece of software designed for a bespoke engine on a completely different, albeit powerful, general-purpose engine – it can be done, but it won't be as fast or as smooth without considerable tuning.
What This Means for the Future of Upscaling
This development, while premature for widespread adoption, opens a fascinating door. For years, AI upscaling technologies like DLSS and AMD's own FidelityFX Super Resolution (FSR) have been largely confined to their respective hardware ecosystems. Nvidia's DLSS, in particular, has been a significant selling point for its RTX cards, offering superior image quality and performance in many titles compared to traditional spatial upscalers or even AMD's FSR. The ability to run DLSS on AMD hardware, even in a modded state, could fundamentally alter the competitive landscape.
If this mod can be optimized to a point where it offers competitive performance and image quality on AMD GPUs, it would diminish one of Nvidia's key hardware advantages. Gamers might find themselves less compelled to choose a specific GPU brand solely for its upscaling capabilities. This could lead to increased competition, potentially forcing both Nvidia and AMD to innovate more rapidly in the upscaling space. For developers, it could simplify the implementation of advanced upscaling, as a single, highly effective solution might become viable across a wider range of hardware. However, the path from a proof-of-concept mod to a polished, widely supported feature is long and fraught with technical and business challenges. Nvidia has little incentive to officially support DLSS on competitor hardware, meaning this likely remains a community-driven endeavor for the foreseeable future.
The Community's Role in Pushing Boundaries
The modding community has a long history of achieving the seemingly impossible, often bridging gaps left by official product development or vendor limitations. This DLSS 5 on RDNA 4 endeavor is a prime example. It demonstrates the ingenuity and persistence of developers who are not bound by corporate roadmaps or hardware restrictions. Their work can serve as a powerful signal to hardware manufacturers about user demand and potential future directions.
While the current 30 FPS on the RX 9070 XT is a far cry from optimal, the fact that it runs at all is a testament to the modder's skill. The eventual goal of matching a 5070 Ti-level performance suggests an ambitious roadmap for optimization. This might involve leveraging specific RDNA 4 features that were not initially accounted for, or developing entirely new algorithmic approaches to tensor core emulation. The success of this mod, however rudimentary, provides a compelling argument for more open standards in AI acceleration or, at the very least, a greater willingness from hardware vendors to explore cross-platform compatibility for their flagship technologies. For now, it remains a fascinating, if rough, glimpse into a potentially more unified future for game rendering technologies.