AMD has unveiled its next chapter in FSR’s evolution, called Redstone. With it, FidelityFX Super Resolution (FSR) is set to take a significant leap forward thanks to the introduction of caching and denoising techniques that should bring Radeon graphics parity with Nvidia’s GeForce solutions.
Building on the AI-powered foundations of FSR 4, Project Redstone represents a noticeable evolution for AMD’s upscaling suite, combining advances in machine learning with new rendering developments to push real-time graphics quality and performance to new heights. Redstone will be based on three main aspects, Neural Radiance Caching, Machine Learning Ray Regeneration, and Machine Learning Frame Generation.
Neural Radiance Caching
Neural Radiance Caching (NRC) is a feature that continually monitors how light bounces in the scene to predict future bounces and store indirect lighting information for later use. This should allow a more consistent and detailed light reproduction, especially on complex scenes with dynamic light sources, all while improving performance. If you have ever selected path tracing in games, you probably have an idea of how heavy it can be, even on the best hardware available. NRC should reduce that overhead.
Machine-Learning Ray Regeneration
Machine-Learning Ray Regeneration is yet another technique powered by neural networks, meant to further boost path tracing simulation speed. It seems identical to Nvidia’s DLSS Ray Reconstruction as it uses AI to reconstruct rays that weren’t fully traced during the rendering pass, while filtering out noisy artifacts in real time. This way, a GPU can simulate far more advanced lighting, with sharper ray-traced reflections and shadows, but without the accompanying calculation cost.
Machine Learning Frame Generation
Machine Learning Frame Generation (MLFG) is perhaps the one that most games will benefit from. Following in the footsteps of FSR 3 Frame Generation, MLFG brings AI to the mix, promising smoother gameplay thanks to its temporal and spatial awareness when generating frames. If the improvements brought by AI on FSR 4 upscaling are any indication, we can expect a noticeable boost in the quality of these generated frames.
Support
FSR Redstone is architected specifically for AMD’s RDNA 4 GPUs, leveraging their enhanced AI capabilities. While the hope of seeing such features on older architectures is still permitted, we wouldn’t hold our breath for it. Even FSR 4 has yet to appear on RDNA 3.
Aside from Multi-Frame Generation – which AMD didn’t mention – Redstone should bring AMD Radeon 9000 GPUs to parity with Nvidia’s GeForce RTX 50 Series. Their inability to correctly run path-traced games should disappear, improving the experience for Radeon owners. An especially important shift, seeing how ray tracing is becoming mainstream.
AMD plans to release FSR Redstone in the second half of 2025, so stay tuned.
