# AMD’s FSR 4: Advanced Path Tracing and AI Supersampling to Rival Nvidia’s Ray Reconstruction
AMD has established itself as a pivotal force in the graphics sector, persistently expanding the limits of GPU capabilities. A notable aspect of AMD’s approach has been its reluctance to implement artificial intelligence (AI) in its FidelityFX Super Resolution (FSR) technology. Even with the advent of FSR 3.1, AMD maintained its focus on temporal upscaling methods, avoiding AI-enhanced features. However, this is poised to shift with the forthcoming launch of **FSR 4**, as the company has announced intentions to incorporate AI-driven elements such as **neural supersampling** and **denoising**.
## AMD’s Renewed Emphasis: Neural Denoising and Supersampling
In a recent article on AMD’s **GPUOpen** developer blog, the firm shed light on its ongoing exploration into AI-driven graphics improvements. The main concentration lies in a **neural denoiser**, designed to refine noisy visuals produced from a limited number of ray samples. This approach is similar to Nvidia’s **Ray Reconstruction** technology, which employs AI to enhance real-time ray tracing by minimizing noise and elevating image fidelity.
Although AMD’s blog didn’t directly reference Nvidia, the parallels between the two technologies were affirmed by **Mateusz Maciejewski**, an AMD engineer, in an interaction on X (previously Twitter). Maciejewski recognized that AMD’s forthcoming feature serves as an alternative to Nvidia’s Ray Reconstruction, igniting enthusiasm among users curious about how AMD’s offering will compare to Nvidia’s.
## Understanding Path Tracing and the Function of Ray Reconstruction
To appreciate the significance of AMD’s upcoming feature, it is crucial to understand the fundamentals of **path tracing** and **ray reconstruction**. Path tracing is a sophisticated algorithm used to emulate **global illumination** within a scene. It entails computing thousands of light rays per pixel, which are subsequently averaged to ascertain the brightness and hue of each pixel. This method is divided into three types of rays:
1. **Primary Rays**: These rays are the first to strike objects in the scene.
2. **Secondary Rays**: These rays calculate indirect lighting by reflecting off surfaces.
3. **Shadow Rays**: These rays discern whether a point on a surface is shadowed.
The synergy of these three ray types establishes path tracing as the benchmark for lifelike lighting in virtual settings. Nevertheless, the computational demands for “true” path tracing are considerable, often requiring hours to render a single frame in films. For real-time scenarios like video games, this level of detail is not practical.
To make real-time path tracing viable, both Nvidia and AMD diminish the number of rays employed in the calculations. However, this reduction introduces **noise** into the image that needs to be resolved. Nvidia’s **Ray Reconstruction** utilizes neural networks to eliminate noise and reconstruct scene details, yielding a high-quality final image. AMD’s strategy will be comparable, but with a notable distinction: **AMD aims to merge denoising and upscaled reconstruction within a single neural network**, which could potentially enhance processing efficiency and improve image quality.
## Open Source and RX-6000 Compatibility: A Glimpse at FSR 4
One of the most thrilling elements of AMD’s forthcoming technology is its **compatibility with older GPUs**, specifically the **RDNA 2 (RX 6000)** series. This marks a significant shift from Nvidia’s strategy, which frequently confines its advanced features to its latest hardware. As stated by **Mike Burrows**, Vice President of AMD’s Advanced Graphics Program, the new path tracing model will be accessible for both **RDNA 2 (RX 6000)** and **RDNA 3 (RX 7000)** GPUs. Naturally, the RDNA 3 series will exhibit superior performance due to its AI accelerators, notably the **WMMA (Wave Matrix Multiply-Accumulate)** function, but the inclusion of RX 6000 GPUs for this feature is a considerable advantage for AMD users.
Burrows also suggested that this novel technology would be launched as an **open-source** solution, akin to earlier iterations of FSR. This open-source mindset has been a cornerstone of AMD’s approach, enabling developers to integrate and refine the technology across various platforms and hardware setups.
### What Implications Does This Hold for FSR 4?
While AMD has not explicitly stated that these new features will be integral to **FSR 4**, all indications point toward that reality. The open-source characteristic of the technology, coupled with its support for older GPUs, positions it as a sensible addition to the next version of FidelityFX Super Resolution. FSR 4 could very well amalgamate **neural supersampling**, **real-time path tracing**, and **denoising** into a comprehensive solution, promising a significant enhancement in image quality.