MIMD hypercube.

MIMD is for Multiple Instruction Multiple Data
Hypercube _ "

from http://www-jics.cs.utk.edu/machines-descrip.html

Intel iPSC/860

128 nodes, 5.1 GFlops peak (MSR ORNL and CS UTK)

Intel iPSC2

64 nodes, (CS UTK)

Processors

The iPSC/860 is a high performance parallel computer system. The processing power of the iPSC/860 comes from its processing nodes. Each node in the iPSC/860 is either a CX or an RX processor. Each CX processor may also be accompanied by an SX scalar processor and/or a VX vector processor. Every iPSC/860 system contains at least one RX node. The CX node is based on the Intel386 microprocessor. An RX node consists of an Intel i860 microprocessor capable of a peak performance of 80 MFLOPS. The i860 has multiple arithmetic units: an integer unit, a floating point adder and a floating point multiplier.

from http://www-jics.cs.utk.edu/machines-descrip.html

The Interconnection Network

The processing nodes of the iPSC/860 are interconnected in a hypercube architecture. In a hypercube of dimension N, each node has N neighbors and the total number of nodes in the hypercube is 2^N. For example, the machine at Oak Ridge National Laboratory (ORNL) has 2^7 = 128 RX nodes. The nodes of a hypercube are assigned unique addresses so that the address of any two neighbors differs in only one binary digit. The maximum data transfer rate between adjacent nodes on the hypercube is 20Mbps.

Memory

The iPSC/860 has a distributed memory architecture. This means that each processor has its own primary memory. The size of memory varies with location. For instance, the machine at ORNL has 8 megabytes of memory at each node. Some of this memory is used by the operating system for internal purposes. Secondary storage is usually available in the form of disks in the Concurrent File System (CFS).

• Welcome to the Intel Scalable Server Products Division web page that contains all the current information on the Intel TeraFLOPS program.