We have studied forced turbulence of compressible magnetohydrodynamic (MHD) flows through two-dimensional simulations with different numerical resolutions. First, hydrodynamic turbulence with Mach number $<M_s >_{rm init} equiv < v >_{rm rms}/ c_s = 1$ and density compression ${< deltarho / rho >}_{rm rms} simeq 0.45$ was generated by enforcing a random force. Then, initial, uniform magnetic fields of various strengths were added with Alfvenic Mach number $<M_A >_{rm init} equiv < v >_{rm rms} / c_{A, {rm init}} gg 1$. An isothermal equation of state was employed, and no explicit dissipation was included. After the MHD turbulence is saturated, the resulting flows are categorized as very weak field (VWF), weak field (WF), and strong field (SF) classes, which have $<M_A > equiv < v >_{rm rms} / < c_A >_{rm rms} gg 1$, $<M_A > > 1$, and $<M_A > sim 1$, respectively. Not only in the SF regime but also in the WF regime, turbulent transport is suppressed by the magnetic field. In the SF cases, the energy power spectra in the inertial range, although no longer power-law, exhibit a range with slopes close to $sim1.5$, hinting the Iroshnikov-Kraichnan spectrum. Our simulations were able to produce the SF class behaviors only with high resolution of at least $1024^2$ grid cells. The specific requirements for the simulation of the SF class should depend on the code (and the numerical scheme) as well as the initial setup, but our results do indicate that very high resolution would be required for converged results in simulation studies of MHD turbulence.