The parent commentator is a bit confused - most of the innovation in these hybrid architectures comes from reducing the computation pressure not just the memory pressure.