Hello, Interval calculations with rounded_arith_opp in Release configuration under msvc 14 result in zero-size intervals. For example consider following: #include <iostream> #include <boost/numeric/interval.hpp> int main(int argc, char * argv[]) { unsigned int ctrl = _control87(0, 0); _control87(_EM_INEXACT | _RC_UP, _MCW_EM | _MCW_RC); using namespace boost::numeric::interval_lib; typedef typename unprotect< boost::numeric::interval<double> >::type interval_t; interval_t test = interval_t(1.) / interval_t(3.); std::cout << "test.lower() != test.upper(): " << std::boolalpha << (test.lower() != test.upper()) << std::endl; return 0; } This test unexpectedly prints: test.lower() != test.upper(): false Tested on x64 platform with /fp:strict and /fp:precise under MS Visual Studio Community 2015 Version 14.0.24720.00 Update 1, Boost 1.60. I've already reported this to https://svn.boost.org/trac/boost/ticket/12018 There you can find some details of what I think is going wrong and proposed workaround patch. I didn't create pull request because I'm not sure if my patch is a correct workaround. Maybe someone will propose more appropriate solution. Best regards, Peter Azmanov.