Access optimization or interleaving at a lower level than linearly mapping DIMMs and channels. x86 cache lane size is 64 bytes, so it must be a multiple. Probably 64*2^n bytes.