TECH NEWS – While the technology is currently only available on the RTX 4000 series, Nvidia’s DLSS lead hasn’t ruled out bringing the frame rate boosting feature to the next generation.
Digital Foundry caught up with Bryan Catanzaro, Nvidia’s vice president of applied deep learning research (aka DLSS), at CES. The interview was conducted after the announcement of DLSS 4. Catanzaro talked about the new transformer models that will replace the current convolutional neural networks in Super Resolution and Ray Reconstruction technology. The technology is smarter, can be trained on a larger data set, and can make better decisions to reduce ghosting and shimmering.
The new Super Resolution model has four times the processing power of its predecessor. Catanzaro didn’t say how much more rendering time it will require, but he says Nvidia believes it will work best on the soon-to-be-released GeForce RTX 5000 graphics cards. The frame generation technology is also being redesigned. They’re ditching the previous model, which was based on the Optical Flow hardware accelerator, and instead using a new model powered by artificial intelligence.
“When we built DLSS 3 Frame Generation, we really needed hardware acceleration to compute optical flow. We didn’t have enough Tensor Cores, and we didn’t have an Optical Flow algorithm that was good enough. We hadn’t developed a real-time optical flow algorithm that ran on Tensor Cores that fit our compute budget. We had the optical flow accelerator that Nvidia had been building for years as an evolution of our video encoder technology, and it’s also part of our automotive computer vision acceleration for self-driving cars.
It made sense for us to use that for DLSS 3 Frame Generation. But the hard thing about any kind of hardware implementation of an algorithm like optical flow is that it’s really hard to improve on. It is what it is, and the failures that came from that hardware optical flow, we couldn’t undo with a smarter neural network until we decided to just replace it and go with a fully AI-based solution, and that’s what we did for frame generation in DLSS 4,” Catanzaro said.
Catanzaro added that porting to RTX 3000 cards is not out of the question. It’s a matter of optimization and engineering, and if it’s implemented, the user experience will be important. The technology, called Multi Frame Generation, will start on RTX 5000 cards and they will see what they can implement on older generations. But along with the higher demands on the Tensor cores, don’t forget that they also perform worse on the older architecture. Catanzaro even stressed the importance of decoupling the updated flip measurement from the CPU to reduce frame time variability (and thus improve frame timing) by a factor of five to ten compared to previous designs. Last but not least, he claimed that Reflex 2 (which is also AI-based) makes gaming feel much better, and he believes that lag-sensitive gamers in particular will love it.
The only question now is what will be accomplished in supporting older graphics cards.
Source: WCCFTech
Leave a Reply