An evaluation of a new standard for floating point precision : A quantitative comparison of Posit and IEEE 754 Float

Abstract: Posit is a new floating point precision representation that has been developed as an alternative to the existing IEEE 754 floating point standard. Prior studies have found several use-cases for posit, but also potential drawbacks. This thesis examines some of the potential drawbacks concerning the precision of posit’s floating point representation. The study is conducted through comparisons of 32-bit posit with 32-bit IEEE 754 for a limited set of operations and algorithms. The result shows that posit is the most accurate at values near one, progressively losing accuracy as the absolute value of the exponent in the number being represented grows larger. Eventually there is a break point where float becomes more precise. The choice of operators or algorithms was not found to have any affect on the precision of the result.