Nvidia Steps up the AI Game with DLSS 3

With the release date of Nvidia’s new range of graphics cards rapidly approaching, the company has made major waves with announcements of new specs and tech. While upping the ante on transistor count is routine at this point, more impressive are the AI improvements in Nvidia’s deep learning super sampling technology (DLSS). With the most recent iteration DLSS 3 only available on the 40 series cards, this new tech could be a real game changer.  

A Growing Industry Standard

AI in gaming is not a recent development in terms of within games, but on an exterior scale, its involvement in the interactive entertainment process has been recently gaining steam. One of the more common contemporary examples can be found in iGaming solutions like Moonrocket. Building e-commerce solutions on a scale like this require a vast amount of input from AI processes to test for bugs and ensure safety, so the resulting platforms are reliable and safe. This service counts AI as the key to its success, and many others take the same route. This level of involvement becomes increasingly necessary the more digital technology evolves. The more moving parts, the more power there is to leverage, and the more potential issues arise. DLSS is one technology that leans on the growing power of hardware and software, and in doing so it raises the potential of what entertainment can accomplish.  

What is DLSS?

DLSS is a term given to a range of technologies that work together to improve game performance and visual clarity. In video games, the higher resolution of the image that is output, the greater the strain on device hardware. This means upping the resolution will decrease the frame rate, even if every individual frame appears clearer. DLSS works by taking a lower resolution image and then using AI upscaling techniques to build a prediction of a higher resolution image. The outputted higher resolution frame is not as accurate as an image that would natively output at that setting, but it’s still superior to lower resolution frames. Since the performance cost of upscaling is less than natively rendering at higher resolution, DLSS is a powerful tool. DLSS 3 takes the process a step further by using AI not just to create additional pixels, but to develop entire frames. This is similar to a technology called frame interpolation, which, as Techdim explains, has been used in videos for some time. If you’ve ever seen a setting on your TV called motion smoothing, that’s effectively the same idea in action.

Demands on AI

The big difference between frame interpolation in games over video is that gaming is much more demanding on hardware. Video from television and online streams tends to operate in the 20-30 FPS range, for example, with typical resolutions of 1080p. It also doesn’t matter if frame interpolation delays video output on television either, as output sound can be matched to compensate. In gaming, users often play in the 60-144 FPS range, at resolutions between 1440p and 4k. Factor in the need for the frames to be displayed as quickly as possible to minimize the disconnect felt by the player, and that the video cards are already doing so much work, and the toll on video hardware is far more intense. Nvidia is no stranger to AI, as we’ve covered at AnalyticsInsight before, but in gaming, DLSS 3 could be their best application yet. With what PCGamer notes as up to 4x frame rate improvement in some games over last-gen tech, gamers are understandably excited. Whether the company can meet demand or manage to fight off the ever-present scalper problem which infects modern hardware, however, those are other questions entirely.