AMD FSR 3 With Fluid Frames Promises 2x FPS Uplift Over FSR 2
AMD made the announcement during yesterday’s RX 7000 series live stream. Version 3.0 Its FidelityFX Super Resolution (FSR). This new version promises to be a DLSS 3 competitor by combining AMD’s Fluid Motion Video frame generation technology with his FSR upscaling. FSR 3 is coming in 2023 and promises up to double the frame rate of FSR 2.
There are very few details about the FSR 3’s inner workings, with AMD only providing rough performance estimates at the time of the announcement. However, AMD hopes to provide more details on his FSR 3 closer to the technology’s release date in late 2023.
But if FSR 3 follows in the footsteps of FSR versions 2 and 1, it could become an open-source frame generation and upscaling platform for all GPUs. FSR has never required proprietary hardware like Nvidia with DLSS, so it’s entirely possible that FSR 3 will feature the same behavior.
The highlight of this announcement is that AMD Fluid Motion Frames technology is used in conjunction with FSR Temporal Upscaling in FSR 3. This specifically shows that FSR 3 actually uses some kind of frame generation technology.
In case you didn’t know, AMD already has a frame generation technology known as AMD Fluid Motion Video. As the name suggests, this technique smoothes the motion of a video by adding frames to it. As a result, we can speculate that Fluid Motion Frames is his 3D rendering version of this same technology, allowing AMD to add frame generation technology to his FSR 3.
However, one thing worth mentioning is that DLSS 3 requires both Tensor Cores. When Optical flow accelerator supported only on 40-series GPUs. So AMD may need a hardware solution to make his FSR 3 work. Additionally, the optical flow accelerator required by DLSS 3 measures motion vectors between two frames, allowing DLSS to render and insert artificial frames between two real frames.
As a result, if AMD wants to take advantage of frame generation, it will need a comparable motion vector solution that can measure motion between two frames. However, I don’t know how AMD will solve this problem. You can take the same route as Nvidia by building a physics motion vector unit directly on the GPU, or you can take the software route with a motion vector algorithm running on the GPU core.
If AMD chooses the former route, FSR 3 could be limited to RDNA3 GPUs or other AMD GPUs with motion vector units inside. But for the latter, AMD can run his FSR 3 on any GPU, as long as the GPU has enough raw horsepower to handle the FSR 3’s computational requirements.