Today’s large computing systems are empowered with high processing capabilities and they are often used to run real time applications. But these systems consume huge amount of energy while executing these applications. In this paper, we have exploited the actual power consumption pattern of a few recent commercial multi-threaded processors and derived a simple power model which considers the power consumption at a coarser granularity instead of finer granularity like DVFS. We have then proposed an online energy efficient task scheduling policy namely, smart allocation policy for scheduling aperiodic independent real time tasks onto such large systems having multi-threaded feature in the processors. We have further added three variations of our proposed policy to efficiently address different situations which can occur at execution time and to further reduce energy for some kinds of applications. We have analyzed the instantaneous power consumption and the overall energy consumption of four proposed task allocation policies along with other five baseline policies for a wide variety of synthetic data sets and real trace data considering different computation time models and deadline schemes. Experimental results show that our proposed policies achieve average energy reduction of 45% (maximum up to 92%) for synthetic data set and 30% (maximum up to 47%) for real data sets as compared to baseline policies. All the proposed policies ensure that no task misses its deadline.