A performance comparison of scalar, vector, and concurrent vector computers including supercomputers for modeling transport of reactive contaminants in groundwater

Sophisticated and highly computation‐intensive models of transport of reactive contaminants in groundwater have been developed in recent years. Application of such models to real‐world contaminant transport problems, e.g., simulation of groundwater transport of 10–15 chemically reactive elements (e.g., toxic metals) and relevant complexes and minerals in two and three dimensions over a distance of several hundred meters, requires high‐performance computers including supercomputers. Although not widely recognized as such, the computational complexity and demand of these models compare with well‐known computation‐intensive applications including weather forecasting and quantum chemical calculations. A survey of the performance of a variety of available hardware, as measured by the run times for a reactive transport model HYDROGEOCHEM, showed that while supercomputers provide the fastest execution times for such problems, relatively low‐cost reduced instruction set computer (RISC) based scalar computers provide the best performance‐to‐price ratio. Because supercomputers like the Cray X‐MP are inherently multiuser resources, often the RISC computers also provide much better turnaround times. Furthermore, RISC‐based workstations provide the best platforms for “visualization” of groundwater flow and contaminant plumes. The most notable result, however, is that current workstations costing less than $10,000 provide performance within a factor of 5 of a Cray X‐MP.