The Modding Frontier: How Enthusiasts are Bringing Nvidia’s DLSS 5 to AMD Hardware

In the rapidly evolving landscape of PC gaming technology, the line between proprietary hardware features and open-platform accessibility is constantly being redrawn by the modding community. The recent debut of Nvidia’s Deep Learning Super Sampling (DLSS) 5, which arrived alongside the launch of NBA 2K27, has ignited a new wave of ingenuity. While Nvidia intends for this advanced AI-driven upscaling technology to remain a hallmark of its own ecosystem, a developer known as danielblnc has successfully engineered a workaround, allowing DLSS 5 to function on AMD’s RDNA 4 architecture.

This development, dubbed "DLSS-NR-on-AMD," represents a fascinating intersection of reverse engineering, neural network translation, and the relentless desire for parity in the PC gaming space. While the performance currently leaves much to be desired, the project serves as a testament to the technical prowess of the modding community and raises significant questions about the future of AI-accelerated graphics.

The Core Facts: What is DLSS-NR-on-AMD?

At its simplest, DLSS-NR-on-AMD is a software wrapper that tricks games into recognizing non-Nvidia hardware as a compatible DLSS 5 environment. By intercepting calls to Nvidia’s NGX API—the proprietary interface that manages DLSS operations—the mod forces the DLSS 5 Dynamic Link Library (DLL) to load on AMD’s RDNA 4 GPUs, such as the Radeon RX 9000 series.

The mod is closed-source, making its exact mechanics a subject of speculation among experts. However, it is widely believed that the tool performs a complex translation of the neural model data contained within the DLSS 5 package. Instead of relying on Nvidia’s proprietary CUDA cores, the mod appears to redirect these calculations through the HIP RT (Heterogeneous-compute Interface for Portability Ray Tracing) backend. This allows the game’s existing FSR (FidelityFX Super Resolution) pipeline to act as a conduit, passing frame data through the neural model to produce the upscaled output.

Key Technical Requirements:

• Target Hardware: Primarily tested on AMD RDNA 4 (RX 9000 series).
• Software Foundation: Requires a DirectX 12-capable title that natively supports FSR.
• The DLL Hurdle: Users must provide a "legally-obtained" copy of the DLSS 5 DLL, as the mod cannot legally distribute proprietary Nvidia files.
• Compatibility Warning: The tool is strictly incompatible with games utilizing robust anti-cheat software, which would flag the API interception as a malicious injection.

A Chronological Perspective: From Leak to Implementation

The journey of DLSS 5 into the hands of AMD users was not a planned feature, but rather a byproduct of software leaks.

The Leak: Shortly before the official release of NBA 2K27, the DLSS 5 DLL was inadvertently exposed online, likely from early test builds. This early access allowed developers to probe the file’s structure and test its compatibility with non-Nvidia architectures.

The Initial Proof-of-Concept: Early testing revealed that simply dropping the DLL into a game folder was insufficient. Developers had to build custom wrappers to handle the translation between Nvidia’s tensor-focused instructions and AMD’s matrix-aware hardware.

The "Cyberpunk 2077" Milestone: Cyberpunk 2077, a title frequently used as a benchmark for graphical fidelity, became the primary testing ground. Initial reports were grim, with the RX 9070 XT struggling to maintain a playable frame rate while running the mod.

The Patching Cycle: Recognizing the performance bottleneck, danielblnc released a series of hotfixes. One update provided a 12% boost in frame rates, moving the game from an unplayable 28 FPS to a "stable" (though still demanding) 30 FPS at 1080p. A subsequent 2% improvement followed, though users in the community have remained critical of the overall stability.

Supporting Data: The Performance Reality Check

It is crucial to manage expectations regarding this mod. While the achievement is technically brilliant, the practical performance is currently in its infancy. In a widely circulated demonstration involving Cyberpunk 2077, the game dropped from a native 80+ FPS to a stuttering 11-12 FPS when the mod was active.

Performance Comparison:

The significant delta between the modded performance and native DLSS 5 performance is attributed to the lack of hardware-level Tensor cores. While both the RTX 50-series and the RX 9000 series support FP8 (8-bit floating point) formats, the execution paths are fundamentally different. Nvidia’s hardware is hard-wired to handle these specific AI tasks with minimal overhead. On the other hand, forcing AMD’s general-purpose compute units to emulate this behavior creates a massive computational tax, leading to the severe performance degradation seen in current benchmarks.

Official Responses and Industry Context

Nvidia has remained largely silent regarding third-party mods, focusing its public messaging on the official rollout of DLSS 5. The company has confirmed that while the feature is currently optimized for the Blackwell (RTX 50-series) architecture, native support will eventually extend to the RTX 40-series, which possesses the necessary hardware for FP8 calculations.

For AMD, the situation presents a double-edged sword. On one hand, the mod demonstrates that RDNA 4 hardware is capable of executing complex AI tasks. On the other hand, the poor performance of the mod highlights the gap between the efficiency of dedicated AI silicon and general-purpose GPU compute.

Industry analysts suggest that the modding community is essentially acting as a "stress test" for AMD’s hardware. If danielblnc or other developers can find ways to optimize the HIP RT translation, it could force AMD to reconsider how it markets its own AI acceleration capabilities in future driver updates.

Implications: The Future of AI Upscaling

The success of DLSS-NR-on-AMD has profound implications for the future of PC gaming.

1. The Death of Platform Lock-in?

For years, proprietary features like DLSS, Ray Tracing, and Frame Gen have been used as marketing levers to drive sales toward specific GPU brands. If the modding community can consistently bridge the gap between these proprietary APIs and open-platform hardware, the competitive advantage of hardware-locked software features may diminish.

2. The Rise of "Hardware-Agnostic" AI

As AI becomes more integral to graphics rendering, the industry is moving toward a standard where software is no longer strictly tethered to the silicon it was designed for. The techniques being used in the DLSS-NR-on-AMD project—wrapping and translating API calls—could eventually become the industry standard for ensuring compatibility across different hardware manufacturers.

3. The Ongoing Challenge of RDNA 3 and Beyond

While the current focus is on RDNA 4, the GitHub repository for the project has acknowledged that RDNA 3 support is being explored. However, the lack of native FP8 support on older architectures makes this a monumental task. The performance would likely be so low that it would serve little more than a technical curiosity, further reinforcing the idea that specialized hardware is indeed necessary for optimal AI-driven performance.

4. The Role of the Modding Community

Perhaps the most significant takeaway is the sheer resilience of the modding community. Developers like danielblnc are not motivated by profit, but by the intellectual challenge of overcoming technical barriers. As Nvidia continues to update its software, these mods will likely improve in tandem. The performance of this mod is the "worst it will ever be"—as optimization continues and the underlying techniques are refined, we may see a day where the gap between native DLSS and modded DLSS becomes negligible.

Conclusion

The emergence of DLSS 5 on AMD hardware via community-led modification is a watershed moment for PC gaming. It proves that software is becoming increasingly fluid, capable of crossing hardware boundaries that manufacturers intended to keep sealed. While the current performance levels prevent this from being a viable solution for the average gamer, the path is clear: the community is moving toward a future where hardware choice is decoupled from software capability.

For now, the project remains a fascinating experiment for those with the technical know-how to navigate GitHub repositories and DLL injections. For the rest of the gaming world, it stands as a reminder that in the world of PC hardware, there is rarely such a thing as a "proprietary" feature—only one that the community has not yet found a way to open. As we look ahead, the collaboration between hardware performance, driver-level optimizations, and community innovation will define the next generation of visual fidelity in gaming.