For a granular flow in hopper in engineering and experimental applications, it is necessary to guarantee the discharge continuously and steadily. The clogging will easily happen if the outlet size is small enough via formation of the arch above the outlet. The clogging phenomenon is also important for studying traffic or evacuation problems. In previous numerical and experimental study, to expedite the experiments or simulations, the perturbations, such as a jet of pressurized air or the vibration of the wall of the hopper, were induced to break the clogging and restart the flow. But these perturbations are hardly normalized and described in modeling the process. In this paper, we present a series of numerical experiments of clogging in the discharge of particles from a three-dimensional hopper through a circular opening. We employ our discrete element method simulation code for large scale dense granular flow based on the graphic processing unit to expedite this simulation. In contrast to pervious studies, here we study the first clogging after opening the outlet of hopper, thus the above perturbations are avoided. From simulating granular flow in hopper in a wide range of outlet size and cone angle, we obtain the size of distribution of avalanche, which is defined as the number of particles that fall through the opening from the outlet opening to the first clogging. The effects of the outlet size and cone angle of hopper on avalanche size are investigated and discussed. The results show that the previous conclusion of the distribution of possibility of avalanche size is also valid in this study. There is a peak in the distribution of possibility of avalanche size, and the distribution can be divided into two regions, which can be fitted with a power-law and an exponential function respectively. The exponential part can be explained by a possibility model which is suggested by Janda et al. From the fitting we find that it has a critical value for the outlet size above which no clogging will occur and the value in this work (4.75d) is slightly lower than in Zuriguel et al.'s experiment (4.94d). Moreover, there is also a critical value for the cone angle of hopper, which supports the inference in previous study and the value in this paper (77) is closed to the predicted one (75) in To et al.'s work.