Abstract: A matrix processor comprises a system with a variable number of buses, each bus having a variable number of processing elements which may operate in parallel. Each bus accesses a port into a memory crossbar and a multiport memory system also accesses crossbar ports. Efficient sharing of bus accesses by processors and synchronization of processors on each bus is accomplished via registers located on the buses, which may be read and written by processors. Interbus synchronization is also accomplished via register accesses. The matrix processor may be configured as a coprocessor or as a stand alone device. A method of synchronizing the processors and buses, performed by at least one processor on at least one bus, includes reading a barrier state of the processors, synchronizing the processing elements on a each bus, reading the barrier state of the buses, and synchronizing each bus.