The amazing part of this is that each individual core makes contact with each region of the (obviously stationary) memory drum once per revolution, providing dynamic cache adjacency.