Inter-Cell Interference Coordination (ICIC) is commonly identified as a key radio resource management mechanism to enhance system performance of 4G networks. This paper addresses the problem of ICIC in the downlink of cellular OFDMA systems where the power level selection process of resource blocks (RB) is apprehended as a sub-modular game. The existence of Nash equilibriums (NE) for that type of games shows that stable power allocations can be reached by selfish Base Stations (BS). We put forward a semi distributed algorithm based on best response dynamics to attain the NEs of the modeled game. Based on local knowledge conveyed by the X2 interface in LTE (Long Term Evolution) networks [1], each BS will first select a pool of favorable RBs with low interference. Second, each BS will strive to fix the power level adequately on those selected RBs realizing performances comparable with the Max Power policy that uses full power on selected RBs while achieving substantial power economy. Finally, we compare the obtained results to an optimal global solution to quantify the efficiency loss of the distributed game approach. It turns out that even though the distributed game results are sub-optimal, the low degree of system complexity and the inherent adaptability make the decentralized approach promising especially for dynamic scenarios.