To be able to set uniform inlet boundary conditions in simulation, there must be an inlet extension at the first guide vane. In the inlet extension, turbulence experiences strong numerical dissipation, which has not been paid attention to. In the current paper, the influence of the numerical dissipation of turbulence on accuracy in predicting heat transfer was discussed. Two cases, where the numerical dissipation of turbulence was neglected, were analysed. In the first case, wrong conclusion about effect of turbulence scale on heat transfer was drawn: blade heat transfer increases with inlet turbulence scale under the same inlet turbulence intensity. The mechanism for the wrong conclusion is that turbulence with larger scale numerically dissipates more slowly in the inlet extension so that turbulence intensity at blade leading edge is greater under turbulence with larger scale, it is the turbulence intensity not turbulence scale itself really affects heat transfer. In the second case, when the numerical dissipation of turbulence is neglected and turbulence parameters at measuring plane of inlet are directly as input for turbulence boundary condition, flow transition is postponed downstream and heat transfer error is greater, however, when the numerical dissipation of turbulence is considered and turbulence parameters at measuring plane are regard as benchmark and matched by adjusting parameters of inlet turbulence boundary condition, the result shows better agreement with experiment. Thus, the correct way to set turbulence boundary condition is to match turbulence parameters at measuring plane by adjusting parameters of inlet turbulence boundary.