Temporal sampling doesn't have to add a frame of latency. You accumulate the _present_ frame with the _past_ frame(s), and any upscaling tech like this has to u

Temporal sampling doesn't have to add a frame of latency. You accumulate the _present_ frame with the _past_ frame(s), and any upscaling tech like this has to use a negligible fraction of the frame time (a few milliseconds at most). This can introduce some ghosting, like in some implementations of TAA, but generally they avoid that by either doing motion blur on high motion-areas or only using the past frame accumulations for low-motion areas.