Accelerated degradation tests (ADTs) are widely used to access the lifetime information of highly reliable products. To obtain the more accurate prediction of lifetime information, how to design an efficient experiment under the limited budget is a critical issue for reliability analysts. Many researchers have addressed this problem and indicated that a two-level design is the optimum strategy for an ADT plan. In addition, most optimum designs were developed under the assumptions that the numbers of measurements and the duration between two inspections within a degradation path are equal for all testing stress levels. Although it is an easier way of operating conditions for experimenters, some real applications showed that the numbers of measurements are different for all stress levels. In this study, based on the Tweedie degradation model, we determine the optimum designs under different operating conditions so that the asymptotic variance of a prediction can be minimized, and compare their performance to the existing strategies.