Conclusion
FSR Redstone is a solid and very welcome upgrade for owners of RDNA 4 based RX 9000-series cards, especially where machine learning is applied to frame generation and ray tracing cleanup. The shift to a fully ML-based frame generation model finally puts AMD on equal footing with NVIDIA in this area. In our testing, FSR Redstone ML Frame Generation consistently delivers image quality that is at least on par with DLSS, and in some scenes it actually looks better, particularly around shadows and fast-moving objects. I guess we can all agree that this is a huge win for AMD, especially considering what previous FSR frame generation offered. Unfortunately, FSR Redstone doesn't seem to offer anything new for owners of previous generation AMD cards, despite community hopes of a compatibility version of FSR 4 ML upscaling officially releasing.
Ray Regeneration is another strong addition. It doesn't radically change how ray-traced reflections or lighting look in a single screenshot, but it noticeably improves resolution, stability and reduces noise, which matters a lot once you start combining ray tracing with upscaling and frame generation. The performance cost is minimal, so there's no reason not to enable it in supported games. I also like that AMD's Ray Regeneration works completely independently of upscaling, whereas NVIDIA's Ray Reconstruction forces you to turn on DLSS Upscaling before you can use it. AMD's approach gives both developers and players much more flexibility in how they structure their rendering pipeline. The biggest problem here is game support—at this time just a single game is supported.
Radiance Caching is the most forward-looking part of Redstone, but also the one I'm still the most cautious about. The idea itself makes sense and fits well with where real-time ray tracing needs to go, but AMD's explanation of the tech was surprisingly poor in the reviewer's guide. The
developer docs are now online and explain things much better. Radiance Caching's biggest appeal is that it's fully real-time, with a continuously learning approach, which avoids the limitations of offline or pre-baked lighting solutions and should adapt better to dynamic scenes. However, documentation and real-world data are still limited, and until games actually ship with the feature enabled, it's difficult to judge how effective it will be or how widely it will be adopted across different types of games.
One area where AMD deserves unambiguous praise is the software side. The driver-level override mechanism to add support for compatible newer FSR versions to games otherwise stuck on FSR 3.1 works extremely well and is an excellent implementation. You no longer have to rely on manual DLL swapping or hoping that a developer updates their integration. Radeon Software handles this automatically, and the in-game overlay with green checkmarks makes it immediately clear which features are active. Good job, AMD.
Game support will ultimately decide how successful Redstone is, but the signs are encouraging. Between native integrations and driver-level upgrades, AMD is clearly prioritizing reach, not just flagship titles. That approach is crucial if features like ML frame generation and ray regeneration are going to become "always on" options rather than novelties you only see in a handful of games.
There are still gaps. Most notably, AMD currently lacks x3 and x4 frame generation modes, and FSR Redstone doesn't seem to bring anything new to those not on the latest RDNA 4 generation, which feels like a missed opportunity given the scale of the Redstone launch and the expectations AMD set going into it. I wouldn't be surprised if higher Frame Generation multipliers arrive next year, but it's something I would have expected to see with Redstone. Even so, as it stands today, Redstone is a meaningful step forward in AMD's efforts to position FSR as a strong competitor to NVIDIA's DLSS.