NVIDIA Keynote Live at SIGGRAPH 2026

· Source: NVIDIA · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Gaming & Interactive Media · Depth: Expert, extended

Summary

NVIDIA's SIGGRAPH 2026 keynote unveiled significant advancements across computer graphics, simulation, and AI, highlighting their convergence. A major announcement was DLSS 5, a new real-time rendering technology that combines traditional rendering with learned generation to achieve unprecedented photorealism. DLSS 5 leverages the rendered frame to enrich appearance while preserving artistic intent, ensuring temporal stability, and operating at real-time speeds via a compact diffusion transformer. It offers extensive developer controls, including model selection, global intensity adjustments, and granular masking for specific scene elements. NVIDIA also showcased progress in AI-driven physics and simulation, demonstrating how AI models trained on simulation data can accelerate complex tasks like climate modeling and aerodynamic design by orders of magnitude. The Cosmos Lab introduced Cosmos, NVIDIA's world foundation model for physical AI, which generates physics-aware synthetic data, environments, and starting points for robotics. Cosmos 3, including a 4-billion parameter Edge model, enables real-time policy execution on devices.

Key takeaway

For game developers and graphics engineers aiming for next-gen visual fidelity, DLSS 5 offers a powerful new lever beyond traditional rendering. You can achieve photorealistic results with granular artistic control, enhancing elements like subsurface scattering and reflections without compromising original intent. For robotics and AI researchers, NVIDIA's Cosmos platform provides a critical data engine for physical AI. You should explore post-training Cosmos models for your specific embodiments or utilize Cosmos Dreams for rapid, physics-aware simulation and policy training.

Key insights

AI, graphics, and simulation are converging to enable next-generation creative tools, digital twins, and physical AI.

Principles

Method

DLSS 5 combines traditional rendering with learned generation, using the rendered frame as a dense, pixel-aligned signal to enrich appearance while preserving artistic intent.

In practice

Topics

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA.