We measure the scale dependence and redshift dependence of 21 cm line emitted from the neutral hydrogen gas at redshift 1 < z < 5 using full cosmological hydrodynamic simulations by taking the ratios between the power spectra of Hi-dark matter cross correlation and dark matter auto-correlation. The neutral hydrogen distribution is computed in full cosmological hydrodynamic simulations including star formation and supernova feedback under a uniform ultra-violet background radiation. We find a significant scale dependence of Hi bias at z > 3 on scales of k 1hMpc −1 , but it is roughly constant at lower redshift z < 3. The redshift evolution of Hi bias is relatively slow compared to that of QSOs at similar redshift range. We also measure a redshift space distortion (RSD) of Hi gas to explore the properties of Hi clustering. Fitting to a widely applied theoretical prediction, we find that the constant bias is consistent with that measured directly from the real-space power spectra, and the velocity dispersion is marginally consistent with the linear perturbation prediction. Finally we compare the results obtained from our simulation and the Illustris simulation, and conclude that the detailed astrophysical effects do not affect the scale dependence of Hi bias very much, which implies that the cosmological analysis using 21 cm line of Hi will be robust against the uncertainties arising from small-scale astrophysical processes such as star formation and supernova feedback.