NVIDIA has been at the forefront of integrating Artificial Intelligence (AI) in gaming technology. Specifically, its Deep Learning Super Sampling (DLSS) has evolved over the years to significantly improve game performance. The latest insights from NVIDIA suggest that DLSS 10 could revolutionize game visuals through full neural rendering interfaced with game engines. This article will elaborate on the evolution of DLSS, its current capabilities, and NVIDIA’s future plans for this technology.
Evolution of DLSS
Initial Phases and the Introduction of Tensor Cores
DLSS started as a technique aimed at leveraging AI to boost game performance. The technology relies on trained neural networks to achieve its objectives. To enable real-time ray tracing and to recover performance lost in the process, NVIDIA integrated Tensor Cores into its GeForce graphics cards, starting with the RTX series.
DLSS 2.0 and 3.0: Quality and Performance
Over time, DLSS underwent several updates. Version 2.0 was a milestone in providing higher quality visuals while maintaining enhanced performance. The subsequent version, DLSS 3.0, added Frame Generation. This feature significantly improved performance, especially in CPU-bound games.
DLSS 3.5 and Ray Reconstruction
DLSS 3.5 continued the trend of innovation by focusing on ray tracing. It introduced the Ray Reconstruction feature, which debuted in the game Cyberpunk 2077 to much acclaim. This feature enhanced the quality of ray tracing under upscaling conditions.
Future of DLSS: Neural Rendering
During a recent 'AI Visuals' roundtable, NVIDIA's VP of Applied Deep Learning Research, Bryan Catanzaro, discussed the potential future of DLSS. He anticipates that future versions, possibly DLSS 10, could fully render games using a neural, AI-based system. This involves the game engine generating information that feeds into a neural network responsible for the complete rendering process. This shift could make games more immersive and visually impressive.