GPU memory model quite different from CPU memory model, with application level explicit synchronization and coherency and so on. I don't think that transparent

GPU memory model quite different from CPU memory model, with application level explicit synchronization and coherency and so on. I don't think that transparent compression would be possible, and even if it would surely carry drastic perf downside