Star formation covers a huge size scale. From the large giant molecular cloud complex of several tens of parsecs to the protostars at the order of solar radius, it is about 9 orders of magnitude in spatial scale. In between, we observe long filamentary dark clouds, large molecular cloud clumps, dense cloud cores that house the formation of protostars, and protostellar accretion disks. Multiply coupled highly non?linear physics are involved in the star formation process, including magnetic field, supersonic turbulence, gravity, radiation and protostellar outflow feedback. Numerical simulation has become a powerful tool to understand and connect these observational snapshots of the entire star formation process. Still, it is a challenge to simulate the entire process in a single simulation. To achieve such a long?term challenge, we need efficient adaptive codes, such as adaptive mesh refinement (AMR) code, that can manage memory usage efficiently and are highly scalable on large?scale high performance supercomputers. In this talk, I present my latest cluster formation simulation results, using our radiative?magnetohydrodynamics AMR code "ORION2", to understand (1) the intertwined filamentary structure of dark clouds, (2) the magnetic properties of the dense cloud clumps, and (3) the protostellar cluster formation in a highly supersonic turbulence region of about 4.6 parsecs in size with the highest resolution of 28 astronomical units. The simulation results match well with latest observations and provide important insight into the process of star formation at different size scales that cannot be easily obtained from observations.