Floats can be problematic even between different IEEE754 architectures, for example some SSE/AVX instructions give slightly different results on AMD and Intel processors. They are only specified with a minimum precision so you can't count on the least significant bits to be consistent.