The usefulness of low/adaptive precision is mainly discussed in deep
learning to reduce computational time and power consumption.
Low/adaptive precision accelerates computations by improving
computational performance per flop or reducing data transfer. In order
to use low/adaptive precision in computer simulations, it is also
necessary to discuss the impact of decreased computation accuracy. This
talk discusses the use of low/adaptive precision on an ICCG iterative
method by evaluating an impact on a convergence rate and reducing
computational time.