Nvidia will teach old GeForce video cards to increase FPS using AI, but then

In an interview with Digital Foundry, Bryan Catanzaro, vice president of research for applied deep learning at Nvidia, said that he does not rule out the possibility of introducing the AI ​​frame generation function for increasing FPS, which has become part of DLSS technology, into older Nvidia video cards in the future. GeForce.

Image source: Digital Foundry / Nvidia

Since its debut in 2018, Nvidia’s deep learning scaling technology (DLSS) has evolved into its fourth version. Its latest iteration switched to a transformer-type AI model, which made it possible to implement a number of new functions, including multi-frame generation (MFG). The latter allows you to create up to three additional frames for each traditionally rendered frame to increase FPS.

Nvidia has been able to implement some new technologies, including DLSS Ray Reconstruction, Super Resolution, and Deep Learning Anti-Aliasing (DLAA) technology, on all GeForce RTX graphics cards since the 20s. th series. However, the first generation Frame Generation, originally introduced as an exclusive feature of GeForce RTX 40 series graphics cards, is not supported by GeForce RTX 30 series and RTX 20 series models. The new multi-frame generator was initially announced only for the latest GeForce RTX 5000.

In a conversation with reporters, Brian Catanzaro noted that he does not rule out the appearance of a frame generation function in older models of Nvidia video cards.

«I think the key here is the issue of design and optimization as well as the end user experience. We are running this frame generator, the best frame generator that is Multi Frame Generation, with 50 series graphics cards. And in the future, we’ll see if we can squeeze something out for the old generation of equipment,” commented an Nvidia representative.

Given Catanzaro’s statement, it can be assumed that the first version of the frame generator may eventually appear on GeForce RTX 30 series video cards. However, it is unlikely that it will appear on GeForce RTX 20 series models. At the same time, most likely, the multi-frame frame generator will remain exclusive to RTX 50 series video cards, since its operation requires significantly more computing power tailored for AI, which these cards provide with new tensor cores.

One of Nvidia’s leading developers also shared some information about the development of DLSS.

«When we created Nvidia DLSS 3 Frame Generation, we absolutely needed hardware acceleration for Optical Flow computing. But we didn’t have enough tensor cores and we didn’t have a good enough Optical Flow algorithm. We did not create a real-time Optical Flow algorithm on Tensor Cores that could fit into our compute power reserves. We had the Optical Flow hardware accelerator, which Nvidia had been building for years as an evolution of our video encoding technology. It was also part of our computer vision acceleration technology for self-driving cars. It seemed like it would make sense for us to use it for Nvidia DLSS 3 Frame Generation as well. But the challenge with any hardware implementation of an algorithm like Optical Flow is that it’s really hard to improve. It is what it is, and the glitches that this hardware Optical Flow caused were things we couldn’t fix with a smarter neural network until we decided to just replace it and move to a completely AI-based solution. That’s exactly what we did for Frame Generation in DLSS 4.”