3 supports 999 digits. remember that 1.23 e2 is 123. e3 is 1230.
floating point double only supports about 15 decimal digits, so in sci notation anything over 15 is starting to be imprecise for doubles. you can go a lot higher if you have big numbers eg e100 where the last 85 digits are all zeros, or tiny numbers with leading zeros. Double supports about 300 +- for that, so 4 digits isnt needed until you go to > 64 bit formats.