This paper addresses the problem of position estimation for salient permanent magnet synchronous motors (PMSMs). It is assumed that stator currents and voltages and rotor speed are measurable signals. Two extensions of the position observer are presented - the first one is based on a standard gradient estimator and the second observer uses recently proposed the dynamic regression extension and mixing (DREM) procedure for adaptive estimation of parameters. Both position observers provide global boundness of all signals. In contrast to standard gradient approach which requires persistent excitation (PE) condition of the regressor, DREM estimator ensures parameter convergence if the regressor is not square-integrable. If the regressor is PE the DREM-based observer (like the one with gradient estimator) ensures exponential convergence of position error to zero value. Simulation results demonstrate higher open loop performance of the position observer with DREM estimator - its transient behaviour is less oscillatory with a faster convergence rate compared to observer with gradient algorithm. Also simulations show that both observers possess some robust properties with respect to parameter uncertainty of the motor.