The modding community has achieved another impressive feat, this time bringing AMD's latest FSR 4.1.2 upscaling models to Intel Arc GPUs with hardware acceleration through their dedicated XMX cores. Better yet, early community testing suggests that the optimized implementation can deliver noticeably better image quality than Intel's own XeSS Super Resolution machine learning-based Temporal Anti-Aliasing and Upscaling (TAAU) solution, all while maintaining remarkably similar upscaling costs.

The breakthrough comes courtesy of the OptiScaler modding community, which has been experimenting with modified FSR 4 INT8 DLLs optimized for Intel's Xe Matrix Extensions (XMX) hardware. While AMD's FSR 4 upscaling models could already run on unsupported GPUs through OptiScaler, taking advantage of Intel's dedicated AI acceleration hardware makes the technology substantially more efficient than relying on conventional shader-based execution.

The timing is particularly interesting, as AMD recently expanded official FSR 4.1 support to older Radeon architectures and integrated graphics through its Adrenalin 26.10.1 driver. The newly available FSR 4.1.2 Lite model — an INT8 model which primarily targets mobile chips, handhelds, and APUs (such as specific Ryzen AI and Z-series/handheld platforms) rather than standard desktop RDNA 2 GPUs — is especially promising for less powerful GPUs, employing a smaller neural network to substantially reduce computational requirements while retaining much of the image quality associated with the full model.