Amplitude analysis is an important tool for the research of the hadron spectrum, in which the maximum likelihood method is used to estimate the parameters of a probability density function. In each optimization step, the likelihood values of a huge number of events from both data and Monte-Carlo simulations are calculated and summed, which is the most time-consuming part of the whole optimization process. We have tried to increase their calculation speed with CUDA and finally found that it is limited by the bandwidth of GPU memory. On the Tesla V100 platform, we nearly reach the maximum limit of the bandwidth utilization, for which our analysis tool is named as Full Bandwidth Amplitude Analysis Software (FALLS). In the application to the high energy physics analysis work, the average consuming time of the whole process is reduced to several minutes while our original program also based on GPU needs several hours.