ABSTRACT
iWarp is a product of a joint effort between Carnegie Mellon University and Intel Corporation. The iWarp component contains both a powerful computation processor and a high-throughput, low-latency communication engine. The architecture and logic designs for iWarp have been completed. In the software area, an optimizing compiler developed for Warp has been retargeted to generate code for iWarp. An iWarp system is composed of a collection of iWarp cells, each of which consists of an iWarp component and its local memory. The iWarp cell, consisting of the iWarp component and local memory, is a building block for a variety of system configurations. A one-dimensional array or ring of a moderate number of iWarp cells, delivering on the order of hundreds of MFLOPS, can be an effective attached processor to a workstation. To balance the high numerical processing capability on the processor, iWarp must be equally efficient in communication.
