Efficient high-accuracy computations in massively parallel systems

An efficient way to avoid the rounding errors is the use of multiprecision and interval arithmetics. In this case, the high accuracy of results and the high rate of problem solution can be reached when a computer system combines the possibilities of the dynamic control of operand length, and parallel data processing. The paper deals with the development of a computing system for high-accuracy computations based on the finegrained SIMD architectures with bit-slice sequential processing. A model reflecting the main features of such processing and an architecture of proposed system are discussed. The some results of numerical experiments are presented to show the possibilities of this system.