AMD’s research suggests it plans to catch up with Nvidia by using neural supersampling and denoising for real-time path tracing.

When you make a purchase through links in our articles, Future and its syndication partners may earn a commission.

Nvidia currently dominates the GPU market thanks to its combination of performance, features and brand recognition. Its advanced artificial intelligence (AI) and machine learning technologies have proven particularly effective, and AMD has not caught up, especially in the consumer market. But the company hopes to change that very soon.

According to a post on GPUOpen, AMD’s research is currently focused on real-time path tracing on RDNA GPUs using neural network solutions. Nvidia uses its own DLSS to upscale images using artificial intelligence, but DLSS has come to mean much more than just “deep learning supersampling”—there’s DLSS 2 upscaling, DLSS 3 frame generation, and DLSS 3.5 ray reconstruction. AMD’s latest research is focused on neural denoising to clean up noisy images caused by using a limited number of ray samples in real-time path tracing – mostly ray reconstruction, as far as we can tell.

Path tracing typically uses thousands or even tens of thousands of ray calculations per pixel. This is the gold standard and typically films often require hours to render a frame. Essentially, the scene is rendered using calculated ray reflections, where even a small offset in the chosen path can result in a different pixel color. Do this many times and collect all the resulting samples for each pixel, and eventually the quality of the result will improve to an acceptable level.

To track a path in real time, the number of samples per pixel must be radically reduced. This results in increased noise as light rays often miss certain pixels, resulting in incomplete illumination requiring noise reduction. (By the way, films also use special noise reduction algorithms, since even tens of thousands of samples do not guarantee perfect results.)

AMD aims to solve this problem with a neural network that performs noise reduction while restoring scene detail. Nvidia’s solution has been praised for preserving detail that traditional rendering takes much longer to achieve. AMD is hoping for similar results by recovering path tracing details using multiple samples per pixel.

What’s new here is that AMD combines scaling and noise reduction in a single neural network. According to AMD, their approach “generates high-quality denoised and supersampled images at higher display resolutions than real-time path tracing rendering resolution.” This unifies the process, allowing AMD’s method to replace multiple denoisers used in the rendering engines, as well as perform scaling in a single pass.

This research could potentially lead to a new version of AMD FSR (FidelityFX Super Resolution) that could meet Nvidia’s performance and image quality standards. Nvidia’s DLSS technologies require dedicated AI hardware on RTX GPUs, as well as an optical flow accelerator to generate frames on RTX 40 series (and later) GPUs.

Current AMD GPUs typically lack AI acceleration features, or in the case of RDNA 3, there are AI accelerators that share execution resources with GPU shaders, but in a more optimized way for AI workloads. It’s unclear whether AMD will be able to run neural network denoising and scaling on existing GPUs, or whether it will require new processing clusters (i.e. tensor units). Achieving this on existing hardware will potentially allow a future iteration of FSR to run on all GPUs, but it may also limit the quality and other aspects of the algorithm.

We’ll have to wait and see what AMD ends up offering. An improved approach to neural path tracing and scaling could provide affordable, high-quality graphics for a wider range of hardware, while keeping the path tracing requirements of games in mind (see: Alan Wake 2, Black Myth Wukong and Cyberpunk 2077 RT Overdrive). , we suspect that AMD will need much faster hardware than existing products to achieve higher levels of image fidelity.