NEC MPI is another new implementation. NEC has experimented with several very different implementations. The one described here is just becoming available on the SX-4 as of this writing and should be standard on the SX-4 in the near future .
NEC MPI is a recent descendant of MPICH, starting from the ch_lfshmem device, which was originally implemented for the SX-4. NEC MPI has been highly optimized for both a single-node SX-4, where MPI uses shared memory for communication, and a multi-node SX-4, where communication between nodes is done through the Internode Crossbar Switch (IXS).
NEC MPI is integrated with the VAMPIR and VAMPIRtrace tools from Pallas, which allows users to visualization message trace information to optimize programs .
Other features of the NEC implementation are at this time limited to what is available in MPICH. Because of its recent release, I have not had an opportunity to assess its usability.