Monte Carlo Simulation data for the CMS experiment can be produced using two software tools. The first, the Full Simulation (FullSim), is a more precise tool based on Geant4 detector simulation. The second, the Fast Simulation, provides a faster but still reliable tool and is based on parametric particle-material interactions. FullSim for the LHC Run-3 shows significant computing performance improvements compared to LHC Run-2. The major modifications of FullSim are the migration from Geant4 version 10.4.3 to 10.7.2, the new software package DD4hep for geometry description, the migration from CentOS7 to the new software platform AlmaLinux8 and using the LTO (Link time optimization) build flag. The challenging CMS detector upgrade plan for HL (High Luminosity)-LHC requires extra efforts due to the increased luminosity and the new and complex detector geometry. FullSim plans to meet the requirements for HL-LHC, which includes continues migration to newer versions of Geant4, the current one is 11.1.2, as well as physics improvements including machine learning (ML) techniques to reduce compute capacity needs. Major progresses of Fast Simulation are reached by a more efficient treatment of the generator particles as they propagate through the detectors. Recent developments include the implementation of an increasing more accurate shower generation, improved track finding and tuning of physics processes. This contribution reports the current Full and Fast Simulation performance innovations and further plans to fulfill the significant higher Monte Carlo Simulation demands in LHC Run-3 and for HL-LHC. Very promising software developments using ML for higher accuracy (Refinement Fast Simulation) and speed up (FlashSim) for the simulation run will be explained.