Don't know what the situation is like today, but not so long ago not all architectures supported floating point numbers. Math operations were emulated in software libraries, which was slow and buggy.