Nvidia currently dominates the high-performance graphics card market, largely thanks to its DLSS technology. However, AMD is actively researching innovative ways to leverage neural networks for real-time path tracing on its RDNA GPUs, a feature currently dominated by Nvidia.
According to a recent blog post on GPUOpen, AMD aims to bring real-time path tracing to its RDNA graphics cards. While Nvidia uses AI accelerators on its RTX cards for DLSS upscaling, AMD is focusing on a different performance enhancement technique: denoising.
Neural denoising in a complex virtual scene.
Path tracing in games like Alan Wake 2 and Cyberpunk 2077 casts a limited number of rays into the scene. In real-time, only a few samples per pixel are cast and bounced, rarely returning to a light source. This results in a noisy image (as seen in the top left of the image above) requiring denoising. AMD is employing a neural network to refine this process.
Nvidia already addresses this with Ray Reconstruction, a powerful yet often overlooked DLSS feature. It significantly improves image quality, preserving details in path tracing that would otherwise require extensive offline rendering. AMD is pursuing a similar approach: reconstructing fine path tracing details from a small number of samples per pixel using a neural network.
AMD’s research, however, combines upscaling and denoising within a single neural network. Their blog post states: “We research a Neural Supersampling and Denoising technique which generates high-quality denoised and supersampled images at higher display resolution than render resolution for real-time path tracing with a single neural network. Our technique can replace multiple denoisers used for different lighting effects in rendering engine by denoising all noise in a single pass, as well as at low resolution.”
This research could be the foundation for the next iteration of AMD’s FSR, potentially rivaling Nvidia’s DLSS in performance and image quality. A key question remains: will this technique require specialized hardware? Nvidia asserts that dedicated accelerators on its RTX cards are essential for DLSS, suggesting AMD might need similar hardware on its GPUs.
It’s possible, however, that AMD could make FSR 4 (or whatever the next version is named) compatible with all graphics cards while still utilizing a neural network. RTX GPUs already have the necessary hardware. Features like Intel’s XeSS demonstrate that running AI models on GPUs through separate instructions is feasible, although often with compromises in image quality and performance.
