The invention discloses an energy consumption management method for a high-performance computing application. Based on a locality principle of a program and a characteristic that large amounts of iterations exist in the high-performance computing application, a running behavior of a high-performance computing job is identified by a tag such as a compute-intensive tag, a memory-intensive tag or the like, and a power consumption state of a system component is adjusted according to the identification tag, so that the purpose of energy consumption management is achieved. The whole method comprises a monitoring step, an identification step, an adjustment step, a prediction step and a feedback step. The monitoring step is in charge of monitoring the running behavior of the job; the identification step is used for performing tag identification on the running behavior of the job in a current monitoring cycle; the adjustment step is used for adjusting the power consumption state of the system component according to an adjustment rule corresponding to the tag; the prediction step is used for predicting a running behavior of the job in the next stage; and the feedback step is used for searching for an optimal adjustment rule suitable for running of the current job according to a running situation of the adjusted job, thereby achieving balance between performance and energy saving.