GPUs will drop memory clocks dynamically, with at least one supported clock speed that's intended to be just fast enough to support scanning out the framebuffer

GPUs will drop memory clocks dynamically, with at least one supported clock speed that's intended to be just fast enough to support scanning out the framebuffer. I haven't seen any indication that anybody is dynamically offlining VRAM capacity.