Oh, you're right! I do remember doing this calculation with integers at some point. I might have been using something like a UUID as my bitfield length.

The context was timestamps -- I (playfully) figured out how many bits we'd need to count Planck times before the expected heat death of the universe, thus being truly future proof. I was expecting the answer to be a rediculous number, but in fact it was more in the "not efficient, but definitely feasible" range.

Floating point gets us there too, of course, but it has its own drawbacks and it isn't nearly as funny.