The head of Nvidia promised that game rendering will never be completely left to AI – AMD disagrees

Nvidia CEO Jensen Huang said that game graphics will never be entirely generated by artificial intelligence – human participation will always remain necessary. AMD did not agree with him.

Image source: nvidia.com

The day after the long-awaited announcement of Nvidia GeForce RTX 50 series video cards, Huang answered questions from journalists. Representatives of the PCWorld resource asked: if today AI is increasingly used to generate and interpolate game frames, will computer graphics be completely generated by AI? The Nvidia CEO responded negatively. The question arose naturally: Nvidia DLSS 3 AI scaling technology involves generating one frame for each drawn by the system, while the new DLSS 4 generates three for each drawn frame. A day earlier, AMD chief gaming solutions architect Frank Azor answered this question in the affirmative. Jensen Huang disagreed.

«Because just remember, when ChatGPT came out, we said, “Oh! Now let’s just generate the book.” But now no one expects this. And the reason is that you need to give it the initial [data]. You need to set them – this is called a condition. Now this is the chat or request context. Before you can answer the question, you need to understand the process. The context could be a [file] PDF, the context could be an internet search. The context can be whatever you specify as the context, right? The same thing happens with games. You need to set the context. And the context for games should be connected not only to the plot, but also to space, the world, and be spatially relevant. So you set the conditions, you set the context, you set the initial pieces of geometry, the initial pieces of texture, and it can generate, create everything else,” Jensen Huang said.

The head of Nvidia further noted: “In ChatGPT, the context is called Rapid Retrieval Augmented Generation [RAG] – the context that routes the text response. In the future, 3D graphics will become generation based on 3D conditions. In DLSS 4, Nvidia’s rasterization engine renders only one out of four future frames. Therefore, out of four frames, 33 million pixels, we draw only two [million]. Isn’t this a miracle? The same thing will happen in games in the future that I just described, it will happen not only with the pixels that we draw, but also with the geometry that we draw, with the animations that we draw, and with the hair that we will render in future games.”

The Nvidia CEO apologized to those for whom his answer was not entirely clear, but concluded that there is and will always be a place for the artist and rendering in games. “But it has taken a very long time for us to now understand that generative AI is indeed the future, but you need to set the conditions, you need to ground yourself with the authors, the artists [and] the intentions,” he concluded.

In other words, video cards will continue to render some part of the frames in the game, although not all of them. But according to Huang, there will not be a complete transition to AI rendering, which AMD believes in; something will always need to be modeled and rendered.