NVIDIA Introduces DLSS Transformer: A New Era of Upscaling with All-RTX Support
NVIDIA Corporation recently officially revealed information about its new model DLSS Transformer, designed to replace the usual convolutional neural networks (CNN) in the task of upscaling for gamers. Previously, the main bet was made on CNN, however, according to the company, this technology has reached a certain “ceiling” in terms of further development. The new system is based on the “transformer” architecture, widely used in modern computer vision and natural language processing tasks. Thus, the transition to DLSS Transformer will improve the clarity and detail of game frames, which is especially important when using ray tracing and high resolutions.
Along with the release of DLSS version 4, NVIDIA has revealed its plans for the graphics lineup RTX 50, lifting the embargo on a number of demo slides. The main announcement is that DLSS-4 will be compatible with the entire RTX lineup, even with earlier series, so many gamers will be able to enjoy the new upscaling capabilities as early as January 30. The only exception will be the RTX 50-exclusive option. multi-frame generation, which will allow more efficient “prediction” of future frames and display an image on the screen with high frequency stability.
Introduction transformer DLSS models take image quality to a new level: NVIDIA emphasizes the improved fight against "flickering" and "halo" (halo effect), which sometimes annoy players with sharp camera movements. However, the company honestly admits that such detail and high precision of processed pixels will require a little more time for calculation, so some users may feel a slight decrease in performance compared to the classic CNN model. But the advantages of the visual component, according to the developers, will be very significant.
To make upscaling even more accessible, NVIDIA will update its NVIDIA application to manage DLSS settings: from the end of January, users will be able to manually choose between CNN and Transformer in all games that support the technology. Previously, to use the new version of DLSS, gamers had to replace DLL files in the game folder, which sometimes led to conflicts between different versions and unexpected bugs. Now these processes will be centrally controlled by one program. This approach will allow experimenting with improvements and selecting the optimal parameters for a specific system configuration. As a result, gamers around the world will be able to choose what is more important to them: maximum speed or maximum quality.
NVIDIA, for its part, is confident that it is developing a vector capable of revolutionizing the field of upscaling. DLSS technology has come a long way since its debut several years ago, gradually increasing the complexity of models and obtaining more advanced machine learning methods. Now, according to the “green team”, the era of transformer networks will be the next round of evolution, opening up even more impressive possibilities for future releases, including upcoming AAA projects with an emphasis on realistic graphics, cinematic effects and a wide range of resolutions.