A distribution is called $m$-grained if each element appears with probability that is an integer multiple of $1/m$.
We prove that, for any constant $c<1$, testing whether a distribution over $[\Theta(m)]$ is $m$-grained requires $\Omega(m^c)$ samples.