Well, not if it's hardware accelerated. Then it's also a physical problem.
Not entirely, however the result is achieved its going to run on the GPU, so its always going to be hardware accelerated technically speaking. The question is how fast it can you do it without resorting to things like customization for 4/8/16/32-bit packed math. AMD may simply be building for a system where the processing can occur using older hardware that has some packed-math support but at a performance penalty in addition to something more on-par with Nvidia's offerings on RDNA2-level hardware.
It does make sense to go this way as basically any sort of upscaling technique is intended as a performance-saver versus native resolution rendering, and as such the lower end the hardware the greater the need for the boost.