DLSS 2.0 still requires a large developer investment, though less than DLSS 1.0, which is why DLSS adoption is still very limited, especially since the vast majority of gamers are not playing at 3840x2160, they're playing at 1920x1080. If you use Steam for your data, over 75% are playing at 1920x1080 or below, and that's a target resolution which is easily made playable by GPUs of the last couple of years without issues, and that's not going to make developers want to sink valuable resources into it. The main benefit of DLSS is vastly improved ray tracing performance, as is being seen with the Minecraft ray tracing beta where 1920x1080 ray tracing performance is unplayable (Source: PCGamer, RTX 2080 used) unless DLSS is.
Now, that being said, with AMD just getting into ray tracing this year with the upcoming RDNA2 cards, and as a proponent of open standards, they will no doubt want to devise not only an agnostic version of DLSS, but one which won't require significant deep learning training by the developer and therefore spur adoption. This technology will be gold in consoles where the reduction in visual fidelity caused by DLSS is hidden by the reduction in detail level on consoles due to weaker GPUs especially as the next generation of consoles will target 8K displays.
Also, don't forget that AMD has had the foundation for what DLSS does for some time inside their FidelityFX package: Contrast Adaptive Sharpening, which supports Dynamic Resolution Scaling, and that's the basics of what DLSS does, rendering at a reduced resolution and displaying at a higher resolution. Of course AMD will not have cards on the market which have AI capability like nVidia's Tensor cores until RNDA2 which will, if the patent applications are followed, allow every USP to act as either a image or AI processor, but AMD is no doubt working behind the scenes on this.
FidelityFX is DLSS's counterpart. NVIDIA has Image Sharpening as well. But they've added this feature later on the NVIDIA Control Panel. Also I wonder the answer too. Because DLSS 2.0 improves the image quality pretty noticeably.
Whether there is a noticeable performance difference with only artificial intelligence, it does not need to raise the image quality excessively, because it restricts both the manufacturers and us only muscle strength. The software should not be forgotten either.
motion interpolation could be even superior to dlss 2.0 to boost fps. console gamers use tvs with motion interpolation to boost fps in console games quite significantly. im asking for this for years, yet unheared. my rx 5700 xt does 19-24fps in the unigine superposition 8k test. with motion interpolation it could be easily 60fps even 144fps.
AMD is light years away from this. They can't even release stable drivers and are too busy releasing USELESS features that don't work
20.4.1 is stable on my box with all AMD hardware