IT people worry about perfection to a single byte, engineer know that no component is perfect and just applies +/- 10-20 percent tolerance. Same deal like with screws. You don't calculate screw to be "just right" because in real world it would be overloaded or tightened too much. Instead you just multiply by safety factor (depending on how critical machine is).
And I thought this thread was a bit dull.
So here, we have a statement that exact numbers in engineering are not NEEDED, because we have to deal with uncertainty. So let's buy a RAM chip with 32KBytes capacity +/-20%. Who cares about exact figures? They are for the birds!
After all, there's "only" a difference of about 2% between 1k and 1K(i). Make that almost 5% between 1M and 1Mi. Now that"s almost 7% between G and Gi. But who cares?
Just use a conservative 80% of the rated capacity of your memories, whatever unit it is given in. You're all set.