As mobile devices are widely used, the performance of their graphics processors has increasingly improved. To meet users’ continuous pursuit of excellent experience, the screen resolution and refresh rate of mobile devices are constantly increasing every year. At the same time, the programmable shading pipeline in mobile games is becoming more complex, which leads to game applications becoming the main source of power consumption for mobile devices. This work studies the rendering pipeline in mobile games and proposes a motion-aware rendering frame rate adjustment method to ensure rendering quality in power-saving mode. Unlike previous prediction models that only consider rendering errors of historical frames, this method builds a nonlinear model between camera pose and inter-frame rendering error and predicts error based on the new frame’s camera pose, thus achieving more accurate frame rate adjustment strategies. In addition, the method also includes a lightweight scene recognition module that can adjust the error threshold according to the specific scene where the player is located, thereby adopting different degrees of frame rate adjustment strategies. Quantitatively compared with the prediction model that only considers historical frame errors, the proposed model improves the prediction accuracy on game frame sequences by more than 30%. At the same time, in the qualitative comparison of user experiments, under the same frame-skipping ratio, the proposed algorithm can achieve higher rendering quality and better user experience. The algorithm integrates historical frame errors and camera information to predict more accurate future frame errors. It also combines prediction and scene recognition results to achieve better dynamic frame rate adjustment performance.