Nvidia introduces dlss 2.0 and promises more image quality

Nvidia announces a second version of its deep learning super sampling technique. Dlss 2.0 offers a higher image quality than the first generation and the artificial intelligence no longer needs to be trained per game.

Nvidia’s dlss technology renders games in a lower resolution and then fills in the missing details with the help of artificial intelligence. The technology works with the Tensor cores of Nvidia’s RTX video cards. With the current implementation, Nvidia has to train an algorithm per game based on game images that have been rendered at a 16k resolution. With dlss 2.0 it is no longer necessary to train a new algorithm for every game.

According to the video card maker, the second iteration of the upscale technique offers more image quality with fewer artifacts. Nvidia claims that in some cases the image looks better than when rendering at a native resolution. With dlss 2.0 it is possible to choose a 1080p resolution and upscale to 4k. According to Nvidia, all resolutions are now available on all RTX GPUs, because the new dlss version performs better. There are three presets for using the technique: Performance, Balanced and Quality.

Game developers need to build in support for the technology; it doesn’t work on all games. In a blog, Nvidia shows examples and videos comparing the two dlss generations using Control, MechWarrior 5, Deliver Us To The Moon and Wolfeinsten: Youngblood.