Even the exact same source code compiled with different compilers, or the same compiler with different compiler options.

Intel Compiler for e.g. uses less than IEEE764 precision for floating point ops by default, for example.