As already mentioned in yesterday’s Big Navi post, AMD is working on a competitor to NVIDIA’s DLSS 2.0 upsampling technology. While we still know very little about it, there are a few educated guesses we can make thanks to the Xbox Series X presentation at HotChips:
For starters, unlike NVIDIA’s DLSS technology, AMD’s solution is going to be OpenSource like the rest of the FidelityFX suite. Secondly, it’s likely going to be based on DirectML which is already a well-established Super-Resolution technique and doesn’t require any additional hardware.
Furthermore, I believe that AMD is going to pair this Super Resolution technology with the rest of its FidelityFX tools such as the denoiser, CAS, optimized AO, and SSR to further improve both image quality and performance.
In terms of quality, just like ray-tracing, the first iteration of AMD’s Super Resolution technology will likely be somewhat inferior to DLSS 2.0, but the added features mentioned above should more or less help alleviate that.
The performance should be in line with traditional upscaling technologies allowing doubling of the input resolution. TensorFlow takes 55ms to upscale a 540p image to 1080p on a 2080 Ti while DirectML takes just 15ms, offering a solid 67 FPS in the process. This should be sufficient to allow the Big Navi GPUs to run intensive games like Far Cry 6 and Godfall at 4K 60FPS+.