Nvidia: AI Upscaling Looks Better Than Native Resolution for Gaming
Gamers were skeptical when Nvidia first unveiled its Deep Learning Super Sampling (DLSS) technology with its Turing GPUs. The question became whether it would be wise, at that point, to trade visual fidelity for increased frame rates. After all, we could always do that by turning down the settings a notch or two. Then Nvidia announced AI-driven frame generation with DLSS 3, and the question came up again, as those “fake frames” looked weird in certain scenarios. That’s all behind us now, according to Nvidia, as it’s claiming DLSS has now surpassed traditional rasterization at native resolution as the best option for maximum visual fidelity.
An Nvidia engineer made the controversial statement in a round table discussion with Digital Foundry. The discussion was part of a campaign by Nvidia and CD Projekt Red to discuss the release of the Phantom Liberty expansion for Cyberpunk 2077, which features the company’s newest version of its technology—DLSS 3.5 with ray reconstruction. Cyberpunk has always been a vehicle for Nvidia’s latest rendering technologies, including the most hardcore ray tracing mode, dubbed Overdrive, and upgraded path tracing technology, which will destroy every modern GPU without using upscaling. Version 2.0 of the game, which is not the expansion but an updated version of the original, includes support for DLSS 3.5. This teed up the discussion with Digital Foundry, Nvidia, and CD Project’s Jakub Knapik.
During the roundtable, Nvidia’s vice president of applied deep learning research Bryan Catanzaro kicked the hornet’s nest by stating, “Cyberpunk 2077 frames using DLSS are much ‘real-er’ than traditional graphics frames. If you think about all of the graphics tricks, all the occlusions and shadows and fake reflections, screen-space effects…Raster in general is just a bag of fakeness, right? So, we get to throw that out and start doing path tracing and might actually get real shadows and real reflections,” according to a transcript of the hour-long chat by Wccftech.
He expanded on this comment by stating, “The only way we can do that is by synthesizing a lot of pixels with AI. It’d be far too computationally intensive to do [path tracing] rendering without tricks. So, we’re changing what kind of tricks we’re using and I think, at the end of the day, we’re getting more real pixels with DLSS 3.5 than without.”
CD Projekt’s Knapik agreed, stating that traditional rasterization is a bunch of hacks, as it involves things like painted lighting and reflections instead of real pixels calculated by path tracing, with DLSS offering “something in the middle.” He said, “In the past, you could say that you were trading off some quality for performance by using DLSS, but with DLSS 3.5, it’s really indisputably a better-looking image than without.”
It would have been awkward for Knapik to disagree with Nvidia’s engineer. Still, most people would agree that DLSS 3 with ray-and-path tracing along with frame generation looks better than native resolution, thanks to more realistic lighting, which includes shadows and reflections.
Native resolution with path tracing and no upscaling would offer the best visual fidelity. However, that combination isn’t possible on modern GPUs and won’t be for many years. According to Nvidia, AI is smarter at rendering PC games more realistically than traditional rasterization. It will improve as time goes by, too, as the AI gets smarter.
CD Projekt’s Phantom Liberty expansion comes out on Sept. 26, but an update to the game arrives today along with new Nvidia drivers that enable DLSS 3.5. We can all see for ourselves what it’s like and whether it is indeed prettier than native resolution.
Nvidia: AI Upscaling Looks Better Than Native Resolution for Gaming | Extremetech
Leave a Reply