The G4DSP-XD3 incorporates four Power Architecture MPC7448/47A compute nodes to provide massive signal and data processing performance. All other data paths including two PMC sites are independently bridged from the AccelerX ring so PCI-X speeds are maintained.
For fast interprocessor messaging, each compute node provides an interface into a central hardware message passing accelerator. This allows for fast exectuion of interprocessor mailbox interrupts and semaphores. Also provided is a common time base allowing messages and/or data buffers to be time-stamped, meaning that each G4DSP-XD3 in the system maintains the local time at the same value.
Multiboard scalability is greatly enhanced by the use of two StarFabric ports at P0. StarSwitch can then be used to interconnect multiple G4DSP-XD3s within the VME environment. The G4DSP-XD3 provides four Gigabit Ethernet ports for code development and debug and system I/O on all four nodes. Software support includes VxWorks, Integrity, LynxOS, BIT and AXIS multiprocessor software including optimized VSIPL and non-VSIPL libraries.