With the release of GeForce RTX 40-series graphics cards, the third-generation Deep Learning Super Sampling (DLSS 3) feature also brings a defining feature of “AI frame generation.” TechPowerUp points out that the Ada Lovelace GPU is able to predict the next frame that is handed over to the GPU to render and generate the frame without involving any graphics rendering pipeline. Interestingly, Nvidia also applies this concept to video encoding to use this “magic” to increase the frame rate of the video.
(Source: Nvidia Technical Blog)
It is reported that the Ada Lovelace GPU’s Optical Flow Accelerator (NVOFA) component can apply the same Optical Flow logic to video applications and graphics rendering.
While predicting the next frame, the GPU can also generate the frame through AI to improve the smoothness of the picture – what Nvidia calls “engine-assisted frame rate conversion” (FRUC).
We’ve seen other forms of “fluidity-enhancing” solutions on some TVs (and AMD graphics drivers) before, but here NVENC also compares two real frames in a video, determines motion vectors, and sets up an optical flow stage to ensure the accuracy of the generated frames.
Finally, Nvidia will release FRUC as a library to help it better integrate into content creation and media consumption applications.
