V. CONCLUSION
In this study, we established an energy model for the embedded GPUs that only requires high-level graphics parameters. The energy model is tailored for the architecture of with tiling and HSR designs specifically for the resource constrained embedded GPUs. Results show that the average energy error rate of the model is only 8.2%. With this energy model, graphics developers can understand how much application-level parameters influence the energy consumption of the program, and can thus adjust the parameters to tradeoff the energy consumption and the quality of the 3D graphic rendering.