Floating Point Precision

From UF HPC Wiki

Problems

Double Precision Floating Point

Assuming your code is double precision floating point, differences in precision in the 14th or 15th decimal place between architectures is not significant.

1. Double precision IEEE FP only gives you about 14 digits of precision (beyond the decimal) to start with. I difference in the 14th or 15th decimal place is not significant.
2. Haku and the phase 3 nodes both use Intel (Xeon family) processors whereas the Phase II nodes are AMD Opterons. So the floating point hardware and software implementation is different. Nonetheless, you are getting the precision inherent in IEEE double precision which is all that you can expect on different hardware.