The algorithm itself has many aspects that aren’t well-suited to traditional supercomputing approaches, including frequent and random data accesses, which plug up I/O, heavy data dependencies as well ...