We analyze the non-asymptotic convergence rate of the multiplicative gradient (MG) algorithm for the log-optimal investment problems, and show that it exhibits $O(1/t)$ convergence rates, in both ergodic and non-ergodic senses.