NVIDIA boasted new DLSS methods in Control and technology prospects

NVIDIA DLSS, machine learning-based full-screen anti-aliasing technology using the tensor cores of GeForce RTX graphics cards, has improved significantly over time. Initially, when using DLSS, noticeable blurring of the image was often observed. However, Remedy Entertainment's new sci-fi action movie Control is certainly the best implementation of DLSS to date. Recently NVIDIA told in detail, how the DLSS algorithm for Control.

NVIDIA boasted new DLSS methods in Control and technology prospects

During the study, the company's specialists found that certain temporal artifacts that were previously qualified as errors can be effectively used to add detail to the image. With this figured out, NVIDIA began working on a new AI exploratory model that used such artifacts to recreate details that had previously fallen out of the final image. With the help of the new model, the neural network began to make huge strides and produce very high image quality. However, the specialists had to work hard to optimize the performance of the model before adding it to the game. The final image processing algorithm made it possible to increase the frame rate by up to 75% in heavy modes.

In general, DLSS works according to the following principle: the game is rendered at several resolutions, and then, based on these pairs of images, the neural network is trained to transform the image in low resolution to a higher one. For each game and for each resolution, you need to train your own model for a long time, so DLSS is usually available only in the most difficult modes (for example, with ray tracing effects), providing acceptable performance in them.

NVIDIA noted that even the new improved version of DLSS still leaves room for improvements and optimizations. For example, when using DLSS at 720p in Control, the flame looks noticeably worse than at 1080p. Similar artifacts are observed in some types of movement in the frame.

NVIDIA boasted new DLSS methods in Control and technology prospects

Therefore, experts are going to continue to improve the machine learning model in order to achieve even more impressive results. And they even showed an early version of their next promising DLSS model using the forest fire scene in Unreal Engine 4 as an example. The new model allows you to restore fine details like coals and sparks, although it still needs optimization in terms of frame rate. When this work is completed, owners of Turing-based graphics cards will receive new drivers with even better and more efficient DLSS modes.



Source: 3dnews.ru

Add a comment