Conclusions

We conclude by remarking that the winning strategy in supercomputer design for the coming decade is certainly going to be acquisition of a software advantage in parallel systems. Essentially all of the MIMD manufacturers are likely to be using competitive hardware. It is clear from recent experiences—for example, Evans & Sutherland—that using anything other than off-the-shelf components is too expensive. It follows immediately that different manufacturers can follow more or less the same upgrade strategies. They will end up with machines with basically similar total bandwidths, reliability, and so on. Thus, the key to success in this market is going to be to develop systems that have the best software. Key points there will be virtual shared-memory environments and general-purpose computing capabilities.