At GTC 2026, NVIDIA presented a talk on neural rendering, focusing on the core computational approaches and showing how the differentiable GPU programming layer dramatically expands the high-performance compute ecosystem.

Neural rendering includes applying machine learning to the final image, DLSS 5 being one such example, but also extends to using ML directly inside the rendering pipeline itself, where small neural networks evaluate materials, decode textures, and compute shading results.

One of the applications of neural rendering is Neural Texture Compression (NTC), an ML approach for a more efficient storage of textures. For example, with NTC, the Tuscan Villa scene shown below can be rendered with 970 MB instead of 6.5 GB of VRAM, without significant changes in quality but featuring a reduced memory footprint.

Neural Materials, another application of neural rendering, was inspired by real-life materials, specifically, how they behave, reflect light, and scatter energy. Neural Materials bring high-quality complex materials to real-time rendering and represent material appearance using learned neural features instead of traditional parameters.

The principle of work is similar to that of the NTC's, and in the showcase scene, a material that normally needed 19 data channels was compressed down to eight, with less time required for rendering.

Omniverse NuRec for autonomous vehicle policies is another application of neural rendering based on different principles of the work. With the technique, you can represent the scene as a field of particles optimized from multi-view images.

You can learn the details in NVIDIA's talk:

Also, subscribe to our Newsletter, join our 80 Level Talent platform, and follow us on Twitter, LinkedIn, Telegram, and Instagram, where we share breakdowns, the latest news, awesome artworks, and more.

Are you a fan of what we do here at 80 Level? Then make sure to set us as a Preferred Source on Google to see more of our content in your feed.