AMD Toyshop Demo showed their Neural Super Sampling and Denoiser technology along with Neural Radiance Caching and ReStir for path tracing on AMD GPUs. All three techniques could be considered "neural rendering" techniques in that they all utilize Ai machine learning models to augment traditional rendering to enhance image quality and performance. I still need to do more research on other "neural rendering" techniques, but the term was only recently made popular by Nvidia. There are however other research papers about different neural rendering techniques that were published years before Nvidia's Blackwell presentation. DLSS upscaling could even be considered a neural rendering technique, but I think people are referring to "RTX Faces" or "Neural Texture Compression" or "Neural Materials" or something. I need to learn a lot more to make a video on that for now. So stay tuned! I hope this vid makes sense for now and explains some questions people might have.
Google Doc used in video with links to every paper at bottom:
https://docs.google.com/document/d/12ZRGkgjG4bqRoiVATO0KIMT8SLKupFn1XqetkbLjH2s/edit?usp=sharing
Digital Foundry's Video Explaining ReStir and Neural Radiance Caching:
https://youtu.be/vigxRma2EPA?si=ROiF65NUkAx0MM7A
Because people seem to like the deep dives, I'm planning on making videos that dive a little deeper and maybe I'll record them at some time other than 3am. Sorry if it's kinda a crappy video because I am improving my workflow these vids are kind of experimental for me.
#rx9070xt #rdna4 #rx9070 #amd #raytracing
Neural Supersampling and Denoising for Real-time Path Tracing
Temporally Stable Real-Time Joint Neural Denoising and Supersampling
Spatiotemporal Variance-Guided Filtering: Real-Time Reconstruction for Path-Traced Global Illumination
ReStir and NRC Nvidia RTX path tracing