High computational cost presents a significant barrier to the general application of coupled computational fluid dynamics and discrete element method (CFD-DEM) simulations, especially so for industrial systems with a large number of particles and complex geometries. In this study, a new cross-platform coupling approach is developed by integrating a CFD solver with a standalone GPU-based DEM solver via network communication. Consequently, the two modelling techniques benefit from the most appropriate hardware architecture. The developed coupling approach shows predictions comparable to experiments on a small-scale fluidized bed. Its computational performance is evaluated on a larger fluidized bed and shows superior performance over the CPU-based parallelization methods, making DEM calculation no longer the computational bottleneck. Its general applicability to handle complex geometrical domains is further demonstrated by simulations of a gas-solid cyclone separator. This work demonstrates the benefits of a novel coupling approach which enables efficient and robust solutions for industrial applications.