AMD FSR 4.1.2 Mod Runs on Intel Arc XMX, Nearly Matches XeSS

AMD FSR 4.1.2 Mod Runs on Intel Arc XMX, Nearly Matches XeSS

AMD’s FSR 4.1.2 neural upscaling technology has been unofficially adapted to Intel Arc graphics cards by the OptiScaler modding community, enabling the modified INT8 model to run on Intel’s Xe Matrix Extensions (XMX) AI acceleration hardware.

Early community benchmarks suggest that FSR 4.1.2 Lite can deliver upscaling performance comparable to Intel’s native XeSS Super Resolution, with some testers also reporting better image reconstruction quality. On an Intel Arc A770, FSR 4.1.2 Lite reportedly achieved 101.5 FPS in The Witcher 3: Wild Hunt Remastered, slightly exceeding the 101.1 FPS recorded with XeSS SR in the same reported test configuration.

The development is significant because AMD’s AI-based upscaling technology is not officially supported on Intel Arc GPUs. The community modification provides an alternative execution path that uses Intel’s dedicated XMX matrix engines rather than relying exclusively on general-purpose GPU shaders.

However, the implementation remains experimental. The reported results come from individual users rather than standardized, independently verified testing, and claims about image quality have not been established across a comprehensive selection of games and configurations.

🔬 How the FSR 4.1.2 Intel Arc Mod Works

#

AMD FSR 4 uses neural-network-based image reconstruction to produce a higher-resolution output from a lower-resolution rendered frame. Instead of relying entirely on conventional spatial and temporal reconstruction algorithms, the technology uses trained models to reconstruct image details.

The OptiScaler community’s adaptation focuses on the INT8 version of the FSR 4.1.2 model, including the reduced-complexity Lite variant.

INT8 inference represents neural-network computations using 8-bit integer values. This can reduce computational and memory requirements compared with higher-precision alternatives, although the actual performance and quality depend on the model, implementation, and hardware.

Intel XMX Acceleration Versus General GPU Shaders

#

Intel Arc GPUs include dedicated XMX matrix engines designed to accelerate suitable AI and matrix-computation workloads.

The modified FSR implementation targets these hardware units to execute the neural upscaling model. This differs from an implementation that relies on general-purpose shader execution or an alternative fallback path.

Using specialized AI hardware can reduce the processing time required for supported neural-network operations. However, the result depends on how effectively the model maps to the GPU’s execution architecture and on the efficiency of the surrounding software.

The important achievement is not that Intel Arc officially supports AMD FSR 4.1.2, but that community-developed modifications can make the model execute through Intel’s XMX acceleration path.

The modification does not establish official interoperability between AMD and Intel graphics software. Compatibility, performance, and image quality remain dependent on the particular mod, driver, game, and GPU configuration.

⚡ Intel Arc A770 Benchmark: FSR 4.1.2 Lite Matches XeSS Performance

#

One of the most notable community tests uses an Intel Arc A770 running The Witcher 3: Wild Hunt Remastered at 1080p with the Quality upscaling preset.

The reported results show FSR 4.1.2 Lite delivering a slightly higher average frame rate and lower upscaling processing time than XeSS SR. The full FSR 4.1.2 model also runs on the modified XMX path, although its processing overhead is higher.

Reported The Witcher 3 Benchmark Results

#

Test configuration: Intel Arc A770, 1080p, Quality preset. Results are community-reported and have not been independently standardized.

The Lite model’s reported frame rate is approximately 0.4% higher than XeSS SR, while its upscaling time is about 11% lower.

These differences are small enough that they should not be interpreted as proof of a general performance advantage. Benchmark variability, scene selection, driver behavior, and test methodology can all influence the result.

Nevertheless, the measurements indicate that the modified FSR 4.1.2 Lite implementation can reach processing times comparable to Intel’s own upscaler in this particular test.

The full FSR 4.1.2 model requires 1.11 ms per upscaling operation, compared with 0.75 ms for Lite. This suggests a meaningful processing-cost difference between the two variants on the tested configuration.

Why the Lite Model Matters

#

A reduced-complexity model can be particularly useful on GPUs with limited processing budgets, including mobile graphics processors and handheld gaming systems.

The objective is to preserve much of the image reconstruction quality of a larger model while reducing the computation required per frame.

That trade-off is important because upscaling adds work to the rendering pipeline. Even when a neural model improves image quality, excessive processing overhead can reduce the frame rate or increase frame-time variability.

If the Lite model retains sufficient image quality while remaining close to XeSS in processing time, it could offer Intel Arc users an additional way to balance visual fidelity and performance.

The extent of that benefit will depend on each game’s rendering characteristics and the specific hardware being used.

📊 Intel Arc B580 Test: Upscaling Latency Nearly Matches XeSS

#

A second community test uses an Intel Arc B580 running Clair Obscur: Expedition 33 at 1080p with the Balanced preset.

Unlike the A770 comparison, the reported results emphasize upscaling processing time rather than average game frame rate.

Reported Clair Obscur: Expedition 33 Results

#

Test configuration: Intel Arc B580, 1080p, Balanced preset. Figures are preliminary community measurements.

The Lite model’s reported processing time is only 0.01 ms higher than XeSS SR. At these values, the difference is small, and further testing would be needed to determine whether it is consistent across repeated runs.

The full FSR 4.1.2 model takes approximately 0.85 ms, making the Lite variant the faster of the two FSR configurations in this particular test.

These figures measure the upscaling stage, not total frame-rendering time. They therefore cannot independently establish overall gaming performance, frame-time consistency, or input latency.

Taken together, the A770 and B580 results suggest that FSR 4.1.2 Lite can operate within a similar processing-time range to XeSS SR on more than one Intel Arc configuration. The small number of tests means that broader conclusions remain premature.

🎮 Image Quality: Can FSR 4.1.2 Surpass XeSS?

#

Beyond performance, some community testers report that FSR 4.1.2 produces sharper or more convincing image reconstruction than the XeSS SR implementation tested in the same games.

Users have also suggested that the Lite model preserves much of the full model’s image quality despite its lower computational cost.

These observations are potentially interesting, but they remain subjective reports rather than results from a standardized image-quality evaluation.

What Needs to Be Tested

#

Upscaling quality varies with rendering conditions, motion, object detail, anti-aliasing, and the quality of the source frame.

A model may reconstruct fine detail effectively in static scenes but produce different results in motion. Common comparison points include:

• Fine-detail reconstruction: Textures, thin geometry, foliage, and other high-frequency image features.
• Temporal stability: Flickering, ghosting, and detail consistency across successive frames.
• Motion handling: Reconstruction around rapidly moving objects, particle effects, and camera movement.
• Visual artifacts: Shimmering, oversharpening, ghost trails, and other reconstruction defects.

A rigorous comparison would evaluate these characteristics across multiple games, resolutions, quality presets, and movement patterns. Ideally, testers would use controlled captures and objective image-quality measurements alongside direct visual comparisons.

Without such testing, it would be premature to conclude that FSR 4.1.2 is categorically superior to XeSS. The current evidence supports a narrower statement: some early testers prefer the modified FSR model’s image quality, while the reported Lite processing times are close to XeSS in the tested scenarios.

For additional context on Intel’s AI upscaling ecosystem, see KAD’s coverage of Intel Arc G3 and the expanding XeSS feature set.

🕹️ Frame Generation Adds Another Dimension to the Results

#

An Intel Arc A770 user also reportedly tested Assassin’s Creed Odyssey at 4K using FSR 4.1.2 XMX upscaling alongside XeSS 4x frame generation, achieving a reported frame rate of up to 120 FPS.

This result should be interpreted carefully because it combines two separate graphics technologies: AMD’s modified upscaler and Intel XeSS frame generation.

Frame generation inserts synthesized frames between traditionally rendered frames, increasing the displayed frame rate without requiring the GPU to render every displayed frame independently.

Consequently, a reported output of 120 FPS with 4x frame generation is not equivalent to 120 FPS of native rendering performance. It also does not, by itself, establish the responsiveness or input latency experienced during gameplay.

The result demonstrates that community modding can combine technologies from different GPU vendors in a single rendering pipeline. It does not yet establish a broadly compatible configuration or a standardized performance advantage.

For additional background on AMD’s neural upscaling development, see AMD’s FSR 4 roadmap and its wider graphics technology strategy.

🧩 Compatibility Limitations and Current Risks

#

The FSR 4.1.2 Intel Arc adaptation remains an unofficial modification, and compatibility is not guaranteed across all games or GPU configurations.

Community reports indicate that some games can experience crashes or other problems. DirectX 12 is currently the primary focus of the reported XMX implementation, while support for other graphics APIs remains more limited.

Official Support Versus Community Modification

#

AMD’s Adrenalin 26.10.1 driver expanded official FSR Upscaling 4.1 support to additional Radeon products, including Radeon RX 6000-series graphics and selected Ryzen AI processors.

That official compatibility expansion does not include Intel Arc GPUs. The Intel implementation relies on community modifications and should not be mistaken for a feature supported by AMD or Intel through their standard graphics drivers.

The modified model and injection path may also be affected by game updates, driver changes, anti-cheat systems, or differences in graphics APIs.

Users should therefore expect experimentation to be necessary and should preserve the ability to restore the original game files and graphics configuration.

For technical details about official Radeon compatibility, consult the AMD Adrenalin 26.10.1 release notes.

Why the Mod Needs More Testing

#

Several areas will determine whether the implementation can become a practical alternative for a wider Intel Arc audience:

• Game compatibility: The injection method and rendering pipeline must work reliably across different engines.
• Driver stability: Updates to Intel graphics drivers or the mod itself could change execution behavior.
• Quality consistency: The Lite model needs evaluation across different visual conditions, not just selected screenshots.
• Performance repeatability: Comparable tests are needed across multiple runs, games, resolutions, and GPU models.
• Frame-generation interaction: Mixed-vendor configurations need separate evaluation of rendering performance, frame pacing, and input latency.

Until these issues are addressed, the mod is best treated as an experimental option for enthusiasts rather than a drop-in replacement for officially supported upscaling technologies.

📝 Conclusion: A Promising Cross-Vendor Upscaling Experiment

#

The OptiScaler community’s work on running AMD FSR 4.1.2 INT8 models through Intel Arc XMX hardware demonstrates the potential for cross-vendor reuse of neural graphics technology.

In the reported tests, FSR 4.1.2 Lite achieved 101.5 FPS and 0.75 ms of upscaling time on an Arc A770, compared with 101.1 FPS and 0.84 ms for XeSS SR. On an Arc B580, its reported processing time of approximately 0.59 ms was nearly identical to XeSS SR’s 0.58 ms.

These results indicate that the modified Lite implementation can deliver competitive upscaling overhead in the tested configurations. Community users have also reported favorable image quality, but those assessments remain subjective and have not been validated through comprehensive standardized testing.

The distinction between unofficial compatibility and official support is essential. AMD’s FSR 4.1.2 technology is not natively supported on Intel Arc through the standard driver stack; the current implementation depends on community modifications and has known compatibility limitations.

If further development improves reliability and expands graphics API support, Intel Arc owners could gain another neural upscaling option alongside XeSS. For now, the project is a promising demonstration of cross-vendor GPU capabilities, with its broader value still dependent on compatibility, image quality, and repeatable real-world performance.