News for hackers. Stuff that matters.

A comment on Making floating point math highly efficient for AI hardware. Parent comment

Re:Making floating point math highly efficient for AI hardware by jhj on Saturday November 10, 2018 @01:09PM (#18422550)
The rounding errors for linear posit are exact numbers (excepting division), assuming higher precision. The rounding errors for LNS add/sub are not exact numbers in the representation in the general case. 2 and sqrt(2) are represented exactly, but (2 + sqrt(2)) is not.
[ Parent | on HN ]

There are no comments on this comment yet.