Multi-precision computations are rarely used in high performance computations. On the one hand, people often believe that is makes no sense to require more precision than what is possible for physical measurements. On the other hand, existing multiple-precision software is much slower than classical numerical software based on machine arithmetic. Changing this situation will take a long time, but in our talk we will outline some reasons for making this happen as well as various technical approaches for making this possible.

Occasion: ICTS-TIFR Discussion Meeting, Bangalore, December 13, 2011

Documents: slideshow, TeXmacs source