Most floating point calculations are only correct according to a standard.

Yes, but who said anything about floating point numbers? I guess I should have explained, but my point was that in "normal" microprocessors the voltage level has nothing to do with the precision of the computation... if it's low enough that errors are introduced the results will be wildly incorrect, not merely imprecise as that statement implies.

Not an option.

http://en.wikipedia.org/wiki/Floating_point#Accuracy_problem...

An annoying problem for distributed/grid systems where the compilers are different across nodes/sites.