The issue you observe with summation “stopping” has been discussed previously at Discourse: Size of long array - #46 by ivanpribec
As @mecej4 shows in that thread, the consecutive integers 2^{24} and 2^{24}+1 both have the same IEEE-32 representation Z’4B800000’ .
I believe it has been mentioned several times on this forum, and in the Intel Fortran Developer forums and documentation, that you should not expect bitwise equal results. (I will update and link the discussions if I manage to find them). The compilers rely on different technologies for the the code generation and optimization leading to differences in the sequence of floating point operations taken, and hence different results:
(The slide is taken from the webinar GPU Offloading: The Next Chapter for Intel® Fortran Compiler given by Intel a few months ago).