The new DLSS 4 Multi Frame Generation function of the brand new RTX Blackwell playing cards has created a state of affairs the place one body might be generated utilizing conventional GPU computation, whereas the following three frames can now be solely generated by AI. That’s a hell of an imbalance, so does one of many individuals liable for making this AI voodoo a actuality suppose we’ll get to some extent the place there are no historically rendered frames?
Jen-Hsun Huang, Nvidia’s CEO, and one of many greatest proponents of AI in nearly damned close to all the pieces, says: no.
During a Q&A session right now at CES 2025 Jen-Hsun was requested the query about whether or not we’re more likely to get purely AI generated recreation frames because the entirety of a recreation pipeline and he was unequivocal in his assertion on the contrary, stating that it is important for AI to be given grounding, to be given context with the intention to construct out its world. In different phrases, AI nonetheless wants one thing to construct from.
In gaming phrases Huang means that it really works in the identical means as we give context to ChatGPT.
“The context might be a PDF, it might be an internet search… and the context in video video games has to not solely be related story-wise, but it surely must be world and spatially related. And the best way you situation, the best way you give it context is you give it early items of geometry, or early items of textures it might up-res from.”
He then brings it again round to DLSS 4 and Multi Frame Generation, and the instance of 1 rendered 4K body and three additional 4K recreation frames.
“Out of 33 million pixels,” says Huang, “we render two [million]. Isn’t {that a} miracle? We actually rendered two and we generated 31. The purpose why that is an enormous deal is as a result of these two million pixels should be rendered exactly and from that conditioning we are able to generate the opposite 31.
“Those two million pixels might be rendered fantastically utilizing tons of computation as a result of the computing that we’d have utilized to 33 million pixels we now channel it straight at two. And so these two million pixels are extremely lovely they usually encourage and type the opposite 31.”
And that is kinda essentially the most beautiful means I’ve ever heard somebody talk about upscaling and body technology, as inspiration for AI-generated pixels. Aww.