en.Wedoany.com Reported - AMD officially announced on May 14 that its machine learning-driven image super-resolution technology, FidelityFX Super Resolution 4.1 (FSR 4.1), will be rolled out in batches to Radeon RX series graphics cards based on the RDNA 3 and RDNA 2 architectures starting July 2026. This decision breaks the technology's previous hardware exclusivity strategy for the RDNA 4 architecture, significantly extending AI-enhanced rendering capabilities to the vast ecosystem of older hardware with a large existing user base.

AMD Vice President and General Manager of the Compute and Graphics Business Group, Jack Huynh, publicly confirmed this timeline through official channels. Radeon RX 7000 series desktop and mobile graphics cards utilizing the RDNA 3 architecture will receive official FSR 4.1 driver integration support as early as July this year; official adaptation for the RDNA 2 architecture (covering RX 6000 series desktop graphics cards, integrated GPUs like the Radeon 680M, and custom chips for handheld PCs such as the Steam Deck) is locked in for early 2027. This move covers a wide range of computing hardware forms, including desktops, laptops, and ultra-low-power handheld devices.

FSR 4.1 is AMD's first image super-resolution solution to fully transition to machine learning (ML) inference, marking a complete restructuring of its rendering strategy from traditional spatial/temporal hand-crafted algorithms to AI neural networks. Utilizing updated inference models, the technology can significantly suppress the flickering and softening of fine geometry in motion scenes, while maintaining structural stability during dynamic resolution scaling in "Ultra Performance" mode. Compared to previous FSR generations, its core upgrade lies in delegating the correction of rendering defects in non-linear dimensions to AI, allowing low-resolution rendering pipelines to output image quality approaching native high resolution.

There are clear differences in the specific execution data precision for FSR 4.1 between RDNA 3 and RDNA 4. The RDNA 4 architecture, with its second-generation AI accelerators, natively supports FP8 (8-bit floating point) compute modes; whereas RDNA 3 graphics cards run INT8 (8-bit integer precision) inference models optimized for older hardware via their first-generation AI accelerators. This explains why support for RDNA 3 and RDNA 2 was not achieved in a single step but required months of intensive engineering validation. New technologies like FSR Ray Reconstruction and FSR Radiance Cache remain exclusive to the RDNA 4 architecture due to their deep reliance on Shader Model 6.6 and will unleash their full potential on high-end GPUs through the DirectX 12 Ultimate pipeline in the future.

The compatibility scope of FSR 4.1 covers a broad range of hardware within the AMD chipset ecosystem. Beyond discrete graphics cards, various embedded platforms equipped with RDNA 2 display cores (such as handheld PCs like the Steam Deck) are also included in the supported list for early 2027, meaning low-power mobile hardware will gain system-level ML super-resolution acceleration capabilities for the first time. The officially recommended Adrenalin driver integration solution will significantly reduce the system compatibility verification burden, eliminating the need for users to rely on third-party tools for untrusted replacement calls.

Currently, the industry's design paradigm for GPU roadmaps is gradually evolving towards a hybrid architecture of "general-purpose shading + dedicated ML tensor acceleration." The extension of FSR 4.1 to older hardware indicates that AMD is beginning to further tap the residual computational value of its first-generation AI accelerators (RDNA 3) and the last generation of pure graphics pipelines (RDNA 2). This technology introduces mature mechanisms from HPC and professional visualization fields—such as low-precision quantized inference, de-ghosting filtering, and sharpening mask deep learning—into the consumer rendering environment, enhancing pixel retention rates and inter-frame consistency in scenarios demanding high computational power.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com