Nvidia largely dominates the list of best graphics cardsand it largely depends on its feature set that has been enabled via DLSS. AMD, however, is not remaining idle. The company is researching new ways to leverage neural networks to enable tracing the path on AMD graphics cards – something that, until now, was only really possible on Nvidia GPUs.

AMD addressed the research in a blog post on GPUOpenstating that the goal is “to move toward real-time path tracing on RDNA GPUs.” Nvidia already uses AI accelerators on RTX graphics cards to upgrade an image via DLSS, but AMD is focusing on a slightly different angle of performance gains: denoising.

When you enable path tracing in a game like Alan wakes up 2 Or Cyberpunk 2077, you only get a small fraction of the projected rays in the scene. In a real-time context, only a handful of samples per pixel are projected into the scene and bounce around, but rarely return to a light source within the scene. This leads to a noisy image – see top left of the image above – which needs to be cleaned up by denoising. AMD applies a neural network to the denoising process.

Nvidia has already covered this technique with Ray Reconstruction, which is a Sorely Underrated DLSS Feature. This makes a huge difference in image quality, preserving path tracing details that would normally take minutes or hours to render for a single offline image. AMD is investigating something similar: taking a small number of samples per pixel and reconstructing the finer details of the path tracing using a neural network.

Get your weekly teardown of the technology behind PC gaming

However, the technique AMD is studying combines scaling and denoising into a single neural network. “We are researching a neural oversampling and denoising technique that generates high-quality denoised and oversampled images at a higher display resolution than rendering resolution for real-time path tracing with a single neural network,” reads -on in the blog. “Our technique can replace multiple denoisers used for different lighting effects in the renderer by denoising all noise in a single pass, as well as at low resolution.”

This looks like foundational research for the next version of AMD’s FSR, which could finally match Nvidia in terms of performance and image quality. The lingering question is whether these techniques require custom hardware. Nvidia says dedicated accelerators on its RTX graphics cards are required for AI-assisted upscaling and denoising with DLSS, so AMD may also need dedicated hardware on its GPUs.

However, there is a world in which AMD could open up FSR 4 – or whatever the next version is called – to all graphics cards while still leveraging a neural network. RTX GPUs already have the hardware, and we’ve seen with features like Intel’s XeSS that it’s possible to run AI models on GPUs via separate instructions, although usually with an impact on image quality and performance.