NVIDIA plans to fully disclose its latest research and advancements in rendering technology, simulation technology, and generative artificial intelligence (AI) at the Siggraph conference, a computer graphics and image technology event held in Denver, Colorado in 2024 from July 28th to August 1st. Siggraph is a top communication platform in the industry that attracts leading talents worldwide to promote the innovative development of graphics and image technology.

This conference will feature over 20 cutting-edge academic papers from NVIDIA's research team, focusing on innovative breakthroughs in synthetic data generators and reverse rendering technology. These achievements provide a solid foundation for training AI models. In the field of diffusion models, the ConsiStory project, a collaboration between NVIDIA and Tel Aviv University, has successfully reduced the time for generating coherent story images from over ten minutes to about 30 seconds by introducing a topic-driven shared attention mechanism. This has brought revolutionary changes to areas such as comic creation and storyboarding.

In addition, NVIDIA has further expanded the application of 2D generative diffusion models to interactive texture painting on 3D meshes, allowing artists to quickly create complex textures based on any reference image. This greatly improves creative efficiency and flexibility. Last year, NVIDIA's achievements in this field won the Best Demo Award at Siggraph's live streaming event. This year, they will return with more new advancements and are expected to once again become the focus of attention.

In the field of physics-based simulation technology, NVIDIA has also made significant achievements. They are committed to narrowing the gap between reality and the virtual world in terms of physical performance, and have showcased the achievements of projects such as SuperPADL through a series of innovative research. The SuperPADL project combines the advantages of reinforcement learning and supervised learning to simulate complex human movements based on textual prompts, and this process can be completed in real-time on ordinary Nvidia GPUs. Another research project uses AI to predict object behavior in the environment, opening up new paths for simulating complex physical phenomena.

NVIDIA has also collaborated with Carnegie Mellon University to develop a new renderer that not only simulates physical light but also has the ability to perform thermal analysis, electrostatics, and fluid mechanics analysis. This brings great convenience to the field of engineering design. At the same time, NVIDIA has made important breakthroughs in hair modeling technology and accelerated fluid simulation.

In terms of rendering technology, NVIDIA has proposed multiple new technologies aimed at improving rendering efficiency and realism. By optimizing ray tracing algorithms such as ReSTIR, they have not only improved the quality of rendered images but also significantly accelerated the speed of modeling visible light and simulating diffraction effects. These technologies have important applications in areas such as simulation training for autonomous driving car radars.

In addition, NVIDIA has showcased a series of versatile AI tools for 3D representation and design. Among them, the fVDB framework has attracted attention due to its optimization for 3D deep learning that matches real-world scale. It provides strong support for building city-scale 3D models, processing large spatial scale and high-resolution data of neural radiation fields, and segmenting and reconstructing large-scale point clouds. NVIDIA has also collaborated with several top institutions to develop an algorithm that can generate smooth and space-filling curves in real-time on 3D grids, effectively shortening the design cycle and improving user experience.