Leaked FSR 4.1.1 drivers confirm INT8 upscaling on RDNA 3.5 hardware

The leak confirms that AMD's AI-based upscaling is technically viable on current-generation handhelds and APUs, despite the company's prior silence on RDNA 3.5 support. For the streaming and cloud gaming sector, this signal suggests that low-power client devices like the Steam Deck successor or ROG Ally series will soon access high-fidelity upscaling previously reserved for discrete GPUs. This reduces the client-side hardware barrier for high-resolution video and gaming. Watch for the official release of AMD's 'lightweight machine learning models' specifically tuned for bandwidth-constrained integrated graphics.

AMD's FidelityFX Super Resolution 4.1.1 leak arrives amid a broader effort by the company to extend AI upscaling across its GPU portfolio. At Computex 2026 in June, AMD's Frank Azor pushed back against reports that FSR 4.1 would not be ported to RDNA 3.5 GPUs, stating that no such decision had been made. The controversy originated from comments by AMD VP David McAfee, and Azor's carefully worded response suggested support for RDNA 3.5 integrated GPUs had not been dropped. The leaked FSR 4.1.1 build in Valve's Proton Experimental effectively demonstrates that the INT8 inference path already functions on RDNA 3.5 silicon, corroborating Azor's implication that the feature remains on AMD's roadmap.

The business and rollout timeline for FSR 4.1 on older hardware has been a moving target. The phased approach reflects the engineering challenge of porting an upscaler originally designed for RDNA 4's FP8 AI accelerators to the INT8 hardware found in older architectures. Digital Foundry noted that the leaked build clearly worked on older models using the less capable INT8 instruction set, and that AMD's phased rollout suggests specific architectural optimizations are being made to shore up performance on earlier-generation cards.

From a technical standpoint, the INT8 model format used in FSR 4.1.1 represents a deliberate trade-off favoring broader hardware compatibility over peak throughput on any single architecture. TechPowerUp reported that the RDNA 3.5 exclusion controversy affects Strix Point, Strix Halo, Krackan Point, and upcoming Gorgon Point and Gorgon Halo APU families, all of which integrate high-performance iGPUs such as the Radeon 890M and Radeon 8060S capable of running the required INT8 operations. For streaming and cloud gaming platforms that rely on client-side upscaling to reduce bandwidth, the confirmation that INT8 works on RDNA 3.5 integrated GPUs means handhelds and thin clients will not need to wait for discrete RDNA 4 hardware to access ML-accelerated upscaling.