Machine type: RISC-based distributed-memory multi-processor.
Models: Paragon XP MP
Operating systems: OSF/1, SUNMOS.
Connection structure: 2-D mesh.
Compilers: Fortran 77, C, ADA.
|Model||Paragon XP MP|
|Clock cycle||20 ns|
|Theor. peak performance:|
|Per Proc. (64-bits)||0.075 Gflop/s|
|Communication bandwidth||350 MB/s|
|No. of processors||64-1536|
Note: and are obtained from coupling various XP MP systems up to a number of 6768 processors.
The Paragon MP is a follow-on to the Paragon XP systems that are around since 1992. It follow the trend that is observed in many systems: combination of a number of processors in one node in a shared memory fashion and combining the nodes via a distributed memory model. In the case of the Paragon MP three i860 XPs are combined on one board. When all processors are involved in computation a maximum speed of 225 Mflop/s\ per node should be attainable. In principle, two processors per node serve as application processors while the third is a message coprocessor. When no communication is going on also this third processor is used for computation.
The memory bandwidth within a node is 400 MB/s while internode communication an aggregate bandwidth of 350 MB/s (175 MB/s in and 175 MB/s out). From release 5.0 on for both the Fortran 77 and C compiler use of the multi processor model is transparant to the user.
The Paragon MP retains compatibility with the Paragon XP/S, XP/E, and the former iPSC/860 systems, an Intel hypercube system preceding them. In particular the the transparent parallel Distributed File System can be used in applications migrated from the iPSC/860. The Paragon has its own parallel file system.