Real-time hybrid simulation (RTHS) combines physical testing and numerical simulation with the same equipment testing capacity, allowing for larger testing specimens and lower scaling error. The scale of the numerical model is determined by the capabilities of real-time simulation. Currently, the central processing unit (CPU) can only simulate small-scale models in real-time. This article establishes an RTHS framework based on Python and graphics processor unit (GPU) computation. Using a soil-structure interaction system as a testing model, the performance of the testing framework is verified by numerical simulation and experiment. The results show that using Python-GPU to speed the unconditionally stable algorithm allows a numerical substructure with 27000 DOFs to be solved in a time step of 20ms. Because the scale of GPU calculation is almost 7 times greater than that of CPU calculation in the same machine, the capability of RTHS is greatly extended.