AMD FSR 4.1.2 Runs on Intel Arc XMX Cores, Matching XeSS Performance With Improved Image Quality
AMD's latest FSR 4.1.2 upscaling technology is now running on Intel Arc graphics cards through dedicated XMX acceleration, thanks to a new community modification developed around OptiScaler. Early benchmarks involving the Intel Arc A770 and B580 demonstrate performance comparable to Intel's native XeSS Super Resolution, while testers report noticeable improvements in image quality, potentially providing Intel GPU owners with a powerful alternative to their existing AI upscaling technology.
The breakthrough comes from the OptiScaler development community, which modified FSR 4 INT8 libraries to utilize Intel's Xe Matrix Extensions (XMX) instead of relying on conventional DP4a instructions. Previous attempts to run FSR 4 on unsupported GPUs suffered from higher processing overhead because the technology could not fully utilize dedicated AI acceleration hardware. By executing the neural network directly on Intel's XMX engines, the modified implementation substantially reduces upscaling costs while maintaining the image reconstruction capabilities of AMD's latest models.
The most promising results come from FSR 4.1.2 Lite, a smaller neural network optimized for devices with limited computing resources, including gaming handhelds and integrated graphics. Community testing indicates that the Lite model can match XeSS performance while maintaining much of the visual quality associated with the full FSR 4.1.2 implementation.
AMD FSR 4.1.2 vs Intel XeSS Performance
| GPU | Upscaling Technology | Average FPS | Upscaling Time |
|---|---|---|---|
| Arc A770 | FSR 4.1.2 Lite XMX | 101.5 | 0.75 ms |
| Arc A770 | Intel XeSS XMX | 101.1 | 0.84 ms |
| Arc A770 | FSR 4.1.2 XMX | 98.6 | 1.11 ms |
| Arc B580 | FSR 4.1.2 Lite XMX | Not reported | 0.59 ms |
| Arc B580 | Intel XeSS XMX | Not reported | 0.58 ms |
| Arc B580 | FSR 4.1.2 XMX | Not reported | 0.85 ms |
Community benchmarks conducted at 1080p. Arc A770 results use The Witcher 3: Wild Hunt Remastered with the Quality preset, while Arc B580 results use Clair Obscur: Expedition 33 with the Balanced preset. These are separate tests and should not be used for direct GPU comparisons.
On the Arc A770, FSR 4.1.2 Lite slightly outperformed XeSS with 101.5 FPS versus 101.1 FPS, while reducing upscaling execution time by approximately 11%. Meanwhile, the Arc B580 demonstrated nearly identical processing times between FSR 4.1.2 Lite and XeSS, recording 0.59 ms and 0.58 ms respectively.
The original Arc B580 community testing also highlighted perceived improvements in image clarity and reconstruction stability. However, these observations remain subjective, and standardized visual comparisons across additional games and resolutions will be necessary to determine whether AMD's implementation consistently outperforms XeSS.
The development follows AMD's expansion of FSR 4.1 to Radeon RX 6000 graphics cards, Ryzen APUs and gaming handhelds, demonstrating the increasing flexibility of machine learning upscaling across different GPU architectures. However, Intel Arc compatibility remains an unofficial modification rather than a feature supported directly by AMD or Intel.
The implementation currently focuses on DirectX 12 games and requires modified libraries alongside OptiScaler. Compatibility varies between titles, with some users reporting instability and crashes, including issues with Crimson Desert. Community experimentation has also demonstrated FSR 4.1.2 running alongside Intel XeSS Frame Generation, although generated frames should not be confused with native rendering performance.
For Intel Arc owners, the availability of another competitive AI upscaler could provide greater flexibility when optimizing image quality and performance. More importantly, the results demonstrate that Intel's dedicated XMX hardware can efficiently execute neural networks developed for competing GPU architectures.
The most interesting aspect of this development is that Intel's own XMX hardware can run AMD's latest neural upscaling technology with almost identical performance to native XeSS. This demonstrates that dedicated AI acceleration hardware does not necessarily need to remain restricted to technologies developed by the same manufacturer.
FSR 4.1.2 Lite is particularly promising for affordable graphics cards and gaming handhelds, where every millisecond of rendering overhead matters. However, the reported image quality advantage needs more extensive testing, especially in scenes involving motion, transparency, fine details and temporal stability.
If GPU manufacturers continue making their AI rendering technologies more accessible across different architectures, gamers could eventually choose upscalers based on image quality and efficiency rather than hardware branding alone. For now, OptiScaler demonstrates what is technically possible, even without official cross vendor support.
Would you use AMD FSR 4.1.2 on an Intel Arc GPU if it delivered better image quality than XeSS, or would you prefer sticking with officially supported upscaling technologies?
