NVIDIA has shared insights into the potential future of Deep Learning Super Sampling (DLSS) using artificial intelligence (AI). Bryan Catanzaro, NVIDIA's VP of Applied Deep Learning Research, during a recent AI Visuals' roundtable, discussed the evolution of DLSS and its envisioned capabilities in its tenth iteration. “DLSS 10 in the far future is going to be a completely neural rendering system that interfaces with a game engine in different ways,” Catanzaro stated, highlighting the possibilities of more immersive and visually stunning gaming experiences.
Evolution of DLSS: From Performance Acceleration to Neural Rendering
Since its inception in 2018, DLSS has undergone significant transformations, with each version aiming to enhance gaming performance through AI and neural networks. The technology, integrated into GeForce graphics cards, has been pivotal in optimizing real-time ray tracing. The recent DLSS 3.5 introduced features like Ray Reconstruction, which have been acclaimed for improving the quality of ray-traced images in games, notably in Cyberpunk 2077. Catanzaro emphasized the gradual shift towards neural rendering, stating, “We're going to be using generative AI more and more for the graphics process.”
Challenges and Opportunities in Neural Rendering
While the vision for DLSS 10 paints a promising picture, achieving full neural rendering interfaced with game engines presents its own set of challenges and opportunities. The traditional 3D pipeline and game engines offer control, allowing teams of artists to build coherent worlds. Catanzaro acknowledged the necessity of these tools, dismissing the notion that AI alone could create complex games. However, he expressed optimism about the role of generative AI in enhancing realism and reducing the cost of developing high-quality gaming environments.
The Road Ahead: Research and Development
The journey towards realizing DLSS 10's capabilities is marked by continuous research and development. NVIDIA is exploring additional neural techniques, such as radial caching and texture compression, which could further expand the DLSS suite. These advancements might necessitate an increase in the number of Tensor Cores in NVIDIA's GPUs
