Lattice Boltzmann method is implemented for conjugate heat transfer of a square cavity with solid obstacles. Natural convection is considered in the fluid region while conduction is assumed in the solid obstacles. By adjusting the velocity field and thermal diffusivities, only one single temperature distribution function is needed to simulate the convection in the fluid region and the heat conduction in the solid obstacles region. As a result, matching boundary conditions at the interface are automatically satisfied. The conjugate heat transfer in a rectangular enclosure of natural convection with solid obstacles is considered next. Streamlines, temperature distributions are presented for different Rayleigh numbers, thermal diffusivity ratios and number of solid blocks. The parallel characteristic of LBM is well fitted to the parallel hardware of graphic processor units (GPU) using a CUDA platform. Implementation of conjugate heat transfer LBM algorithm on GPU is presented next. It is found that GPU can accelerate the computation by a factor up to 20 as compared to the non-parallel CPU code. It is demonstrated that lattice Boltzmann Method is an effective approach to simulate conjugate heat transfer with solid obstacles. Published by Elsevier Ltd.