Recently I've become involved in multiple projects where inherent limitations in floating point numbers are leading to problems (for example, carry out a few hundred matrix multiplications and the ...