GDC 2025 Session Viewer

This session explores the strategic management of dynamic lighting in global illumination via machine learning. It highlights how MLP compression significantly reduces storage requirements while maintaining visual quality and mitigating light leakage. The discussion also emphasizes the efficiency of running inference on Tensor Cores with CUDA and seamlessly mapping results to textures. Furthermore, the session delves into the use of H265 compression for distant GI globalmaps, balancing minimal precision loss and optimized performance. It also addresses workflow optimization through hardware-accelerated ray tracing and adaptable training methods, which enable rapid content production and immediate previews for artists. Attendees will gain insights into how these advancements streamline game development processes and enhance visual rendering.

Takeaway

Attendees will gain insights into using machine learning for enhanced global illumination, specifically how MLP compression reduces storage while maintaining visual quality. They will also learn about employing H265 video compression for distant views. Furthermore, they will discover how to implement rapid training and baking processes via hardware ray tracing and integrate machine learning with runtime rendering to optimize visual fidelity and efficiency in game development.

Intended Audience

Technical artists, programmers, and researchers who are keen on exploring the intersection of rendering and machine learning