Free energy perturbation (FEP) is the approach to predict the binding strength between the candidate molecules and their biological target. FEP allows calculation of the ligand-binding free energies by constructing a series of non-physical intermediate states connecting the bound and unbound states. In addition, FEP allows calculation of relative binding free energies between different ligands. We have implemented the FEP method on the cloud computing platform (XFEP) to evaluate the binding affinity between candidate molecules and their biological target at scale, from which false positives can be filtered out before wet lab experiments. Our XFEP technology is based on enhanced sampling algorithms, statistical analysis methods, and our proprietary XForce Field. As confirmed by both retrospective and prospective testing results, the predicted values can be strongly correlated with experimental data, and the average prediction error of XFEP in R-group replacement and scaffold-hopping calculations is about 1.0 kcal/ mol. Powered by our integrated technology platform and cloud-based GPU acceleration, we are able to perform calculations with the XFEP technology for hundreds of drug candidate compounds within a day and effectively reduce the cost of calculation.