If you are looking for such accuracy (where 50_000_000.00006 is different from 50_000_000.06) then you need a specialised library and do very careful coding. I

If you are looking for such accuracy (where 50_000_000.00006 is different from 50_000_000.06) then you need a specialised library and do very careful coding. If you are averaging bazillions of numbers and they range from weeny to huge and you thing 50_000_000.000006 != 50_000_000.06 for your purposes then it is hard for me to see that you are doing anything sensible. This is, except in very specialised cases that I cannot imagine, a non issue. Just add them up and divide, (and check for overflow, that could ruin your day!)