Numerical simulations of physical phenomena are the usual use cases for HPC architectures. However, as they produce more and more data, how can we efficiently analyze and visualize their output ? In this talk, we will present ParaView as a framework for visualizing and analyzing extreme scales of scientific data. Focus will be given specifically to data distribution and resource allocation along with the "in-situ" workflow and current support for heterogeneous compute architectures. Strong of 20 years of development with performance in mind, ParaView is continuously improved, relying on state of the art libraries (such as MPI, OpenMP, VTK-m).