Yes I mean results/ output numbers. U mean I should upload code ? Pl provide email id. I will send it to u

This has nothing to do with Ubuntu, but that floating point numbers are rounded differently by different processors, and that different compilers can execute floating point operations in different orders because the optimizations are different.  

Depending on the specific calculations, round-off errors can grow arbitrarily large.  A famous example is the N-body gravitational problem, N>2, where numerical errors grow exponentially in time.

I agree Prof Daniel. but my concern is how to identify which is giving a reliable results.... there should be a mechanism where results obtained from 32 bit and 64 bit match. thats what i want to do. pl help 

In principle 32 bit or 64 bit operating systems do not change the floating point calculations.  What is making a difference is whether  in the Fortran program you use 32 bit or 64 bit floating point numbers (single or double precision, REAL*4 or REAL*8). 

Of course double precision calculations should be more precise that single precision.  Sometimes quadruple precision is available, but is much slower to compute.  It can be used to check double precision calculations, as double precision is used to check single precision.

Dear Colleagues,

It is known to check of real numbers for equalities, it is necessary to compare them with some accuracy. Give a fragment of the source code.

BW

Alexander.