AMD FSR 4 'Redstone' Brings Machine Learning to Graphics Upscaling

AMD has officially unveiled FidelityFX Super Resolution 4 (FSR 4), codenamed 'Redstone,' marking a significant departure from the company's previous upscaling approaches. For the first time, AMD is embracing machine learning-based upscaling, directly challenging NVIDIA's DLSS technology that has dominated the AI-powered graphics enhancement space since 2018.

The Shift to Machine Learning

FSR 4 represents a fundamental architectural change for AMD's upscaling technology. Previous versions of FSR (1.0 through 3.1) relied on spatial and temporal algorithms that analyzed pixel data and frame history to intelligently upscale lower-resolution images to higher resolutions. While effective and compatible across a wide range of hardware, these approaches couldn't match the image quality that machine learning models could achieve.

With FSR 4, AMD is leveraging dedicated AI accelerators present in its RDNA 4 architecture and newer GPUs. The technology uses neural networks trained on vast datasets of high-quality gaming imagery to predict and generate pixel data with unprecedented accuracy. This approach allows FSR 4 to reconstruct fine details, reduce artifacts, and maintain temporal stability better than algorithmic methods.

The machine learning model at FSR 4's core has been trained on diverse gaming scenarios, including challenging cases like fast motion, particle effects, transparent surfaces, and fine geometric details like hair and foliage. This training enables the system to make intelligent predictions about what pixels should look like at native resolution, even when working from much lower-resolution input.

Performance and Image Quality Gains

Early demonstrations of FSR 4 show impressive results. In AMD's internal testing, games running at 1080p with FSR 4 upscaling to 4K achieved image quality comparable to native 4K rendering while delivering 2-3x performance improvements. The performance gains vary by title and quality preset, but consistently outperform FSR 3.1's already-respectable numbers.

Image quality improvements are particularly noticeable in motion. FSR 4's temporal stability reduces the shimmer and ghosting that occasionally plagued earlier FSR versions, especially on thin geometry and specular highlights. The technology also handles transparent effects and particle systems more gracefully, maintaining detail and reducing break-up artifacts.

AMD has introduced multiple quality presets for FSR 4:

• Ultra Quality: Minimal performance gain but maximum fidelity, ideal for competitive gaming
• Quality: Balanced mode offering strong visuals with notable performance improvements
• Balanced: Moderate quality with substantial performance gains
• Performance: Maximum frame rate boost with acceptable image quality
• Ultra Performance: Extreme performance mode for very demanding scenarios

Each preset adjusts the internal rendering resolution and applies different model configurations to balance performance and quality according to user priorities.

Hardware Requirements and Compatibility

Unlike previous FSR versions that worked across multiple GPU generations and even competitor hardware, FSR 4 requires specific hardware capabilities. The technology is exclusive to AMD GPUs with RDNA 4 architecture and newer, specifically those with sufficient AI acceleration hardware.

This hardware requirement represents a strategic shift for AMD. While the universal compatibility of FSR 1.0-3.1 was a major selling point, AMD has evidently decided that the image quality advantages of machine learning justify the narrower hardware support. The company is maintaining FSR 3.1 for older hardware, ensuring existing GPU owners still have access to competitive upscaling technology.

For developers, FSR 4 integration requires plugin updates and some optimization work, but AMD has designed the API to be familiar to those who've worked with previous FSR versions. The company is providing extensive documentation, sample code, and direct technical support to encourage rapid adoption.

Competing with NVIDIA DLSS and Intel XeSS

FSR 4 enters a competitive landscape where NVIDIA's Deep Learning Super Sampling (DLSS) has established itself as the gold standard for AI-powered upscaling. DLSS 3.5, the current version, combines upscaling with frame generation and ray reconstruction to deliver exceptional image quality and performance.

Comparisons between FSR 4 and DLSS 3.5 show the technologies trading blows depending on the scenario. DLSS maintains advantages in some edge cases, particularly with ray-traced reflections and complex transparency effects, but FSR 4 demonstrates superior handling of fast motion and certain particle effects. Real-world performance differences are often negligible, with game-specific implementation quality mattering as much as the underlying technology.

Intel's XeSS (Xe Super Sampling) offers another competitor, using machine learning acceleration on Intel Arc GPUs while maintaining fallback support for other hardware. FSR 4's image quality generally surpasses XeSS, though Intel continues improving its technology through regular updates.

Developer Adoption and Game Support

AMD has been working closely with game developers to ensure strong FSR 4 support at launch. Several major titles have already confirmed FSR 4 integration:

• Multiple AAA releases scheduled for 2025 will feature day-one FSR 4 support
• Existing games with FSR 3.1 can potentially be updated to FSR 4 through patches
• Game engines including Unreal Engine 5 and Unity are receiving official FSR 4 plugins

The development community has generally responded positively to FSR 4, appreciating AMD's decision to prioritize image quality while maintaining reasonable integration complexity. However, some developers have expressed concerns about maintaining multiple upscaling technologies (DLSS, FSR, XeSS) in their titles, increasing testing and optimization workload.

Technical Innovation: Frame Generation Plus

Beyond upscaling improvements, FSR 4 introduces 'Frame Generation Plus,' an enhanced version of AMD's frame generation technology. This feature uses AI to predict and insert interpolated frames between rendered frames, effectively multiplying frame rates.

Frame Generation Plus offers several improvements over the previous implementation:

• Reduced latency through optimized prediction algorithms
• Better handling of UI elements and HUD overlays
• Improved artifact reduction in high-motion scenarios
• Support for higher base frame rates (now works effectively above 60 FPS)

When combined with FSR 4 upscaling, Frame Generation Plus can deliver dramatic performance improvements. A game running at 1080p with FSR 4 upscaling to 4K and frame generation enabled can achieve frame rates 3-4x higher than native 4K rendering, making high-refresh-rate 4K gaming accessible on midrange hardware.

The Broader Implications

AMD's embrace of machine learning for FSR 4 reflects broader industry trends. AI acceleration is becoming increasingly important across computing, from content creation to gaming to professional applications. GPU manufacturers are dedicating more die space to AI accelerators, recognizing that machine learning capabilities are essential for competitive products.

For consumers, FSR 4 makes high-resolution, high-refresh-rate gaming more accessible. Gamers can achieve 4K 120Hz experiences with more affordable hardware, and those with high-end GPUs can push into 4K 240Hz territory that would be impossible with native rendering.

The technology also benefits VR gaming, where maintaining high frame rates across dual high-resolution displays is particularly challenging. FSR 4's combination of upscaling and frame generation can significantly improve VR performance and visual quality.

Conclusion

FSR 4 'Redstone' represents AMD's most ambitious graphics technology release in years. By embracing machine learning and dedicating hardware resources to AI acceleration, AMD has created an upscaling solution that competes directly with NVIDIA's market-leading DLSS while maintaining the company's commitment to developer-friendly implementation.

While the hardware requirements limit FSR 4 to newer AMD GPUs, the image quality and performance improvements justify this constraint. For gamers with compatible hardware, FSR 4 delivers a transformative experience, making high-resolution, high-refresh-rate gaming more accessible than ever.

As game developers integrate FSR 4 into their titles and AMD continues refining the technology through driver updates, the gap between AI-upscaled and native rendering will continue narrowing. We're entering an era where the question isn't whether to use upscaling, but which upscaling technology delivers the best experience for each specific game and hardware combination.

AMD's willingness to fundamentally rethink FSR demonstrates the company's commitment to competitive graphics technology. FSR 4 doesn't just catch up to the competition—it establishes AMD as a serious contender in AI-powered graphics enhancement, setting the stage for continued innovation in how we render and display the games of tomorrow.