Perhaps in a few years when we'll start to see things like CXL providing cache-coherency between main memory and GPU memory this will become important?