The folding mechanisms of membrane proteins are notoriously hard to determine, due to the multiple events involved in the folding process. In recent years, the development of single molecule techniques has opened the door to studying individual folding events experimentally. However, even in these single molecule experiments the structural details underlying the observed transitions can only be inferred. Similar to E. coli as a model organism, rhomboid protease GlpG is typically used to study membrane protein. Previous single molecular experiments have suggested that GlpG has an anomalously low thermodynamic stability. By performing molecular simulations and energy landscape analysis, we showed that the seemingly low stability was due to the presence of folding intermediates. Our finding was confirmed by a subsequent experimental study by the same experimental group, where our predicted intermediates were observed. On the technique side, we developed our next generation simulation package: OpenAWSEM and Open3SPN2. OpenAWSEM achieves orders of magnitude of speedup with GPU compared with single core CPU, and enables rapid prototyping force fields with automatic derivative calculations.