Apparently the floating point Math functions do not return the same results on all platform / engines / architectures.
At this point we should probably also verify that all operations are actually fully deterministic on the same environment.
Some pointers: