The nCUBE is a message passing hypercube MIMD architecture. Its memory is distributed over the processors and there is no global memory. All the communication takes place through the sharing of messages between the nodes.
The system uses custom build processors that work at 20 MHz and they are rated at 7.5 MIPS, 3.3 MFLOPS (single precision) and 2.4 MFLOPS (double precision). Each node can have from one to 64 MB of main memory. The configuration used to collect timing data had 128 processors, each with 4 MB of RAM.
A small UNIX kernel runs on all the nodes and in addition to normal kernel functionality it supports special communication and synchronization tasks. The system provides global operations function calls (such as global summation), that cause transparent communication of the nodes. Message passing functions and general synchronization functions are also provided. I/O is supported from both on the SUN host and the nCUBE file systems.
The hypercube architecture of the nCUBE gives it special communication
flexibility. A 
node hypercube can simulate 
hypercube with no noticeable lost in communication performance making
it perfect for clustering algorithms.
The nCUBE, as others message passing machines, has a limited buffer for incoming messages, thus it is very important to synchronize data receiving rate with data sending rate. If the buffer limit is exceeded lost of data occurs.