This letter investigates the uplink achievable spectral efficiency (SE) of massive MIMO systems, where the base station (BS) utilizes the linear minimum mean square error (LMMSE) detector and is deployed with the uniform rectangular array (URA). The novel expressions on the achievable SE have been derived by considering a finite dimensional physical channels model and assuming the BS has perfect channel state information (CSI). Based on the derived analytical expression, we further derive the closed-form expression on the total achievable SE for a purely line-of-sight (LoS) channel and explore the effect of the physical parameters on the total achievable SE. Results show that the total achievable SE tends to a saturation rate as the number of BS antennas and/or the average transmit power increases under a purely LoS channel. For a finite dimensional channels, the total achievable SE can not be improved by increasing the number of the paths. Instead, we take advantage of the sparsity characteristics of the finite dimensional channels and propose a beam selection scheme by select the dominant eigenvalues of steering matrix. Results show that the proposed scheme can achieve near-optimal performance with massive MIMO system.