NVIDIA CEO Jen-Hsun Huang released the latest generation of AI graphics technology DLSS 5 at the NVIDIA GTC conference on Monday, aiming to make video game graphics more realistic while reducing computing power consumption. This new system combines structured graphics data from traditional 3D graphics pipelines with generative AI models that can "predict" and "fill in" screen details, allowing the GPU to generate detailed scenes and lifelike characters without having to render every element from scratch.
Jen-Hsun Huang said in his keynote speech that DLSS 5 is the product of merging "controllable 3D graphics, real benchmarks and structured data for virtual worlds, with generative AI and probabilistic computing." In his view, one end is predictive calculation based entirely on physical and geometric rules, and the other end is probabilistic generation with a high sense of reality. The combination of the two allows developers to create content that is both “beautiful and stunning” and highly controllable.
He pointed out that this concept of fusing structured information with generative AI will not just stay in the gaming industry, but will be repeated in one industry after another. In Huang's discussion, structured data is the "foundation of trustworthy AI," and DLSS 5 is the embodiment of this concept in the field of real-time graphics.
Although the gaming business accounts for a smaller portion of Nvidia's current revenue compared to its historical peak, it is gaming that has allowed the company to grow into what it is today. Huang framed DLSS 5 as a concrete example of a broader computing paradigm shift, suggesting that this technology path will extend into areas such as enterprise computing that extend well beyond gaming.
When talking about enterprise-oriented applications, Huang Renxun cited as examples enterprise data platforms such as Snowflake, Databricks and BigQuery as representatives of structured data sets that can be utilized by future AI systems. He believes that in the future, AI will analyze and reason about these data structures at a speed far exceeding that of humans. All kinds of intelligent agents will use both structured databases and unstructured "generative databases". The latter represents "the vast majority of the information in the world" in his words.
In Huang Renxun's vision, this technical paradigm based on structured data and superimposed with generative AI capabilities will drive a series of industries from games to enterprise software to reconstruct their computing methods, and DLSS 5 is a cutting-edge case of this trend.