AMD is set to reveal its next-generation upscaling technology at GDC 2022 later this month. Likely tailor-made for its upcoming Radeon RX 7000 “RDNA 3” GPUs, we’re going to see the successor to FSR 1.0 which saw widespread adoption due to its novel implementation. Although a simple spatial upscaler paired with a sharpening filter at heart, it enjoyed a fair bit of success on account of the wide range of devices supported.
With both NVIDIA and Intel touting their advanced AI-based (temporally fed) upscalers, it’s about time AMD came up with a comparable next-gen alternative. It’s worth remembering that AMD helped develop Epic’s TSR upscaler used in Unreal Engine 5. As such, it’s almost a certainty that it’ll incorporate temporal feedback in its next-gen upscaler.
What remains to be seen is whether AMD will make use of machine learning or some sort of dedicated matrix units on the GPU to speed up the process. NVIDIA already uses its Tensor cores for the same, and Intel plans on leveraging its XMX cores (of which there are plenty on the Arc Alchemist GPUs). There have been numerous reports of AMD leveraging a machine learning chip on its Radeon RX 7000 “RDNA 3” GPUs, but whether these materialize or not remains to be seen.
PS: Keep in mind that Intel’s XeSS upscaler will be made open-source, and will work on NVIDIA and AMD hardware capable of mixed-precision compute. We expect AMD’s solution to be open-source as well (as part of its OpenGPU initiative).