NVIDIA at SIGGRAPH 2026: DLSS 5, Cosmos 3 Edge, and the Convergence of Neural Rendering and Physical AI

The keynote also featured live demonstrations of broader neural rendering research, illustrating how these techniques are changing how digital environments are built for creative tools, industrial design, and simulation .

Cosmos 3 Edge: Open World Models for Local Physical AI

NVIDIA introduced Cosmos 3 Edge, a new open world model specifically built for local physical AI on edge GPUs . It is a 4-billion-parameter omni-model that can understand and generate text, image, video, ambient sound, and action . Cosmos 3 Edge was made openly available at SIGGRAPH, bringing frontier world models to edge devices — including Jetson, RTX PRO, and GeForce RTX platforms — for real-time inference in robotics and autonomous systems . It is the latest addition to the Cosmos 3 family, which also includes Cosmos 3 Super and Cosmos 3 Nano .

Newton Physics Engine and Industry Partnerships

NVIDIA showcased the Newton physics engine, an open-source, GPU-accelerated differentiable physics engine co-developed with Google DeepMind and Disney Research . Built on NVIDIA Warp and the OpenUSD standard, Newton is designed for robotics simulation and training, enabling high-performance, physically accurate simulations on GPUs . The engine was first announced at GTC 2026 and continues to be a central part of NVIDIA's physical AI stack alongside Isaac Lab and Cosmos World Models .

The keynote emphasized the convergence of graphics, simulation, and agentic systems. NVIDIA expanded its Omniverse libraries and Agent Toolkit to bring physical AI capabilities into existing applications . A dedicated Physical AI Day ran on July 21, with sessions on world models, robotics, and digital twins featuring partners across industrial and autonomous systems .

SIGGRAPH 2026 Research Papers

NVIDIA's Real-Time Graphics Research lab presented multiple SIGGRAPH 2026 conference-track papers, including:

• VideoNeuMat: Neural Material Extraction from Generative Video Models
• Toward Richer Material Generation via Procedural Data Enhancement
• Taming optimization variance in compact neural shading networks

The broader SIGGRAPH 2026 program also featured NVIDIA contributions on 3D Gaussian splatting, autonomous vehicle simulation technologies (NuRec, OmniDreams), and capture-to-simulation pipelines for in-the-wild 3D scenes .

Taken together, the announcements at SIGGRAPH 2026 signal NVIDIA's strategy to weave neural rendering, open world models, and physics simulation into a single platform — one that serves game developers, roboticists, and industrial designers alike.