Online portfolio selection is important in the field of quantitative investment． Ｒecently，the emerging investment opportunities call for portfolio selection strategies that can be efficiently computed． However， online portfolio selection strategies with high performance often require exponential or polynomial computation， which hinders their practical applications． This paper proposes a novel universal portfolio selection strategy: the“Sub-Gradient Projection”( SGP) ． This paper first applies the SGP idea to portfolio construction and gains the rebalance rule． Its competitive performance is analyzed theoretically，which shows that SGP is a universal portfolio selection strategy． Moreover，SGP strategy needs linear computational time，which is quite efficient． Empirically，SGP strategy is back-tested using the datasets from US and China． The results show that SGP strategy can achieve a similar cumulative wealth equivalent to that of the latest strategies with much shorter computational time． The experiments on parameter sensitivity show that SGP is insensitive to its parameters． Moreover，it can sustain reasonable transaction costs．