Nvidia CEO’s Support for DLSS 5 Challenged by Employee Remarks

Earlier this week, Nvidia introduced DLSS 5, touted as a groundbreaking AI-driven visual upscaling technology. It claims to utilize a game’s color and motion vectors for each frame, enhancing scenes with photorealistic lighting and materials. However, the gaming community’s response was largely negative, with many dismissing it as just another AI-generated filter.

During a live event later in the week, Nvidia CEO Jensen Huang pushed back against this criticism, asserting that the perception of DLSS 5 as mere “post-processing at the frame level” is misguided. He emphasized that the technology offers a more nuanced approach than a simple filter that alters the final 2D image based on broad internet data.

Yet, new insights from Nvidia’s marketing specialist Jacob Freeman seem to challenge Huang’s assertions. When asked by PC gaming YouTuber Daniel Owens if DLSS 5 uses a single 2D frame along with motion vectors to create the output frame, Freeman confirmed, “Yes, DLSS 5 takes a 2D frame plus motion vectors as an input.” He elaborated that DLSS 5 is trained to comprehend complex scene elements like characters, hair, fabric, and translucent skin, as well as various lighting conditions, all by analyzing just one frame.

For those less versed in technical jargon, the contradiction here is significant. This directly opposes Huang’s earlier statement from March 17, where he claimed, “It’s not post-processing, it’s not post-processing at the frame level, it’s generative control at the geometry level.” He insisted that game developers have direct control over the process, distinguishing it from generative AI, which he referred to as “content-control generative AI” and labeled it as neural rendering.

In essence, the Nvidia representative is suggesting that DLSS 5 operates as a generative AI filter that references a single image, while Huang insists it utilizes comprehensive data, including 3D geometry.

To sum it up, as Owens points out, DLSS 5 appears to be merely applying a filter over a screenshot. This has led to further backlash online, with many accusing Huang of misrepresenting DLSS 5’s capabilities in his recent comments. This isn’t the first time he’s faced allegations of misleading consumers.

It seems that DLSS 5 may not be extracting any additional information beyond that single frame. This could explain why some lighting effects in the initial demonstration appeared subpar; DLSS 5 is relying solely on an image of the lighting to generate new visuals. Ultimately, it looks like DLSS 5 is more akin to an upgraded version of a standard generative AI filter rather than a revolutionary geometry-level rendering technology.