Nvidia’s DLSS 5 uses generative AI to enhance realism in video games, with ambitions beyond gaming

Nvidia CEO Jensen Huang used the company’s keynote at Nvidia GTC on Monday to introduce DLSS 5, a new version of the chip maker’s AI graphics technology designed to make video games more realistic while using less computational power.

The new DLSS 5 combines traditional 3D graphics data with generative AI models that can predict and fill parts of an image, allowing Nvidia GPUs to produce detailed scenes and lifelike characters without rendering each element from scratch.

“We combined controllable 3D graphics, ground truth of virtual worlds, and structured data… with generative artificial intelligence and probabilistic computing,” Huang said during his keynote. “One is completely predictive, the other is probabilistic but very realistic.”

Combining these two ideas — structured data with generative AI — allows developers to create content that is “beautiful, stunning, and manageable,” Huang said.

“The concept of integrating structured information and generative AI will replicate itself in industry after industry,” Huang said. “Structured data is the foundation of trustworthy AI.”

Gaming makes up a smaller portion of Nvidia’s revenue today than it has historically, even though this is the industry that made Nvidia what it is today. Huang framed the DLSS 5 approach as an example of a broader shift in computing, suggesting that the approach could extend beyond gaming and even into enterprise computing.

The billionaire CEO pointed to enterprise data platforms such as Snowflake, Databricks, and BigQuery as examples of structured data sets that future AI systems could analyze and generate insights from.

“In the future, what will happen is that these data structures will be used by AI, and AI will be much, much faster than us,” Huang said. “Agents of the future will use structured databases as well as an unstructured database, the generative database. This database represents the vast majority of the world.”