There are two formulas for SD. You need to use the n-1 version which gives an unbiased estimate of population sigma. These are still uncertainties with smaller samples because the sigma estimate, while unbiased, won;t be exactly the real sigma. The t distribution is designed to take care of that.