Power consumption is an increasing concern in real-time systems that operate on battery power or require heat dissipation to keep the system at its operating temperature. Today, most processors allow software to change their frequency and voltage of operation to reduce their power consumption. Frequency scaling in real-time systems must be done in a way that ensures that the tasks' deadlines are met. In this paper we present the Growing Minimum Frequency (GMF) algorithm for voltage and frequency scaling in uniform multiprocessors for real-time systems. This algorithm runs in polynomial time and computes the optimal voltage and frequency assignment, achieving better power efficiency than previous algorithms. We present the optimality proof and evaluate the practical improvement over previous algorithms with simulated tasksets. Our evaluation shows up to 30% power efficiency improvement over previous algorithms.