Unlocking DLSS 5: Creative Solutions from the Modding Community

When we recently evaluated the performance of DLSS 5 across the new RTX 50 lineup, the results were less than encouraging. For instance, the RTX 5090 plummeted from a robust 192 FPS to a mere 76 FPS at 4K resolution, while the RTX 5060 Ti fell from 129 FPS to a disappointing 46 FPS at 1440p.

However, enabling upscaling provided only a slight 9% boost in performance. This can be attributed to DLSS 5’s neural rendering, which processes frames after upscaling, thereby increasing its workload in accordance with the monitor’s resolution rather than the render resolution.

Remarkably, within just a fortnight, the modding community has devised two innovative methods to tackle this issue, albeit with some limitations. The first method seeks to reposition the neural processing earlier in the rendering sequence, while the second proposes transferring the task to a separate graphics card entirely.

Neural Pass Goes First

The first of these solutions is Neural Upstream, a ReShade add-on that can be installed via the DLSS5-Autopilot tool. This approach reorganizes the traditional rendering pipeline. Instead of the standard sequence of rendering > DLSS Super Resolution > neural rendering, it suggests a new order: rendering > neural rendering (at the render resolution) > DLSS Super Resolution > final output.

By enhancing a smaller image, the neural network utilizes significantly less data, thus preserving frame rates. At 4K, with DLSS set to Quality mode, the model processes 2.7 million pixels compared to the standard 8.3 million. Impressive results have been documented; for example, users reported that on an RTX 4080 playing Assassin’s Creed Shadows at 4K, the costs dropped from around 32 FPS to 13 FPS, offering an efficiency boost of approximately 61%. Similar improvements have been noted in other popular titles like Stalker 2, Cyberpunk 2077, and GTA V Enhanced.

However, a crucial caveat exists: by feeding the neural network less image data, the output quality may potentially diminish. Nvidia’s decision to place this processing step at the end likely ensures higher fidelity, as the current DLSS 5 already grapples with fidelity concerns.

Reassigning Resources: Dual-GPU Processing

The second innovative technique comes from Marcelo Guibout, who developed MGPU Bridge, another ReShade add-on that allows neural rendering to take place on a secondary GPU. By executing the neural pass on a separate card while the primary card handles the rendering, it effectively reduces the workload on the main GPU.

In tests conducted using two RTX 5060 Ti cards at 1080p in The Blood of Dawnwalker, the results were illuminating:

• DLSS Mode with Neural Rendering on Render GPU: Dramatically reduced frame rates by 35-60%.
• Neural Rendering on Second GPU: Minimal impact on frame rates, often between 0% to 17%, while also keeping the primary GPU cooler.

However, this method does come with its own set of challenges. The MGPU Bridge demands two GPUs and two separate monitors, meaning the neural output from the second GPU must be displayed through the card that executed it. This setup is reminiscent of dedicated PhysX cards from earlier generations but may make it impractical for many gamers.

Expanding Compatibility: Creativity in the Face of Limitations

Alongside these performance enhancements, the modding community has also been hard at work on compatibility. The DLSS Unlocked project combines several tools to facilitate neural rendering and multi-frame generation across RTX 20, 30, and 40 series GPUs on both Windows and Linux. While the implementation may not follow Nvidia’s official pipeline, it allows wider accessibility for gamers eager to experiment.

Caution is warranted, especially with DLL injections. It’s crucial to avoid using these modifications in games with anti-cheat systems. Additionally, always source DLL files from trusted places to avoid potential malware threats.

Conclusion

In a world where game performance continues to evolve, the creative efforts of modders seeking to enhance DLSS 5 illustrate the passion and ingenuity at play in the gaming community. While Nvidia’s official approaches may fall short on occasion, these alternative methods provide new avenues for gamers to explore enhanced performance without compromising image quality. As these solutions develop, we can expect an exciting future for graphical fidelity and performance in gaming.