Three scenarios for constructing a PetaFLOPS computer were presented by the Architecture Working Group. The first was a network of supercomputers (e.g., Cray-type machines), the second was a network of high-performance, massively parallel workstations (e.g., Intel Paragon), and the third was a fine-grain, active memory machine. From the standpoint of establishing interconnect and memory requirements, these three architectures can be summarized in terms of their degree of parallelism and their grain size in the following manner: