AMD FSR 4.1: RDNA 3.5 Architecture May Be Excluded from Support

In recent weeks, AMD announced the extension of support for FSR 4.1 technology to graphics cards based on previous architectures, specifically the Radeon RX 7000 (RDNA 3) and Radeon RX 6000 (RDNA 2) series. However, according to the latest rumors, this backward compatibility may not extend to APUs with integrated graphics based on the RDNA 3.5 architecture, effectively excluding the Radeon 890M, 880M, 860M, and 840M models found in the Ryzen AI 300 series processors.

The release timelines indicate that the implementation for RDNA 3 will not be ready before July, while for the RDNA 2 architecture, we will have to wait until early 2027. This temporal delay is due to the need to optimize the algorithm: RDNA 4 chips are natively much faster, and AMD engineers had to work to ensure that execution in INT8 mode (intended for prior generations) could keep pace with and maintain the visual quality of the native FP8 counterpart.

According to internal sources (David McAfee, Corporate VP and General Manager of AMD's Client division), the inclusion of RDNA 3.5 in the FSR 4.1 support plan is currently not planned. The company is assessing the pros and cons, but the current inclination appears to lean towards a "no." This caution is based on precise technical barriers.

The first is that FSR 4 utilizes neural networks that rely directly on the AI accelerators (WMMA units) of the RDNA 4 architecture. Although iGPUs share the same architectural base, they have a significantly lower number of compute units and a reduced power budget. The execution of AI upscaling could consume enough of the integrated GPU's resources to negate the performance gain from the upscaling itself.

Another aspect is the memory bottleneck. A desktop architecture like the Radeon RX 9070 XT can benefit from over 600 GB/s of bandwidth. In contrast, processors from the Strix Point or Krackan Point families must share system memory (DDR5 or LPDDR5X) between the CPU cores and the graphics section. Since AI-based upscalers require additional memory accesses to manage motion vectors, historical buffers, and the neural network itself, the demand for bandwidth risks disproportionately saturating the system.

Finally, there is also a user experience factor: owners of solutions like the Radeon 890M or 880M often run titles with conservative quality settings or at low resolutions. In such scenarios, the visual benefit of an advanced AI upscaling model is less perceptible compared to what would occur on a high-end GPU operating at 1440p or 4K.

Although the final decision has not yet been formalized, leaving a glimmer of hope for support for RDNA 3.5, the current scenario outlines a complex situation for the next-generation handheld PC and ultrathin laptop market, which may not benefit from AMD's latest AI innovations. It's no coincidence that RDNA 4 has been discussed in recent months.

UPDATE

Despite it being quite clear that FSR 4.1 support on RDNA 3.5 is "uncertain" and not "excluded," AMD, through Frank Azor, Client and Graphics Marketing of the company, has decided to clarify the following: "I would like to clarify that no decision of the type reported and implied in this context has been made. At this time, we are unable to provide further details on any future projects related to the products. We continue to listen to our customers and take your opinions into account."