A solution for rendering massive scenes on multiple GPUs will be presented. This new method analyzes the memory access pattern of the path tracer and defines how scene data should be distributed among GPUs with minimal performance loss. This new out-of-core mechanism has been implemented in Blender Cycles. Cycles has been further extended with technologies (e.g. OpenMP, NVIDIA CUDA with unified memory support, Message Passing Interface) that support running on supercomputers and can take advantage of all accelerators allocated on multiple nodes. Another extension that will be presented is the creation of a bridge between the client computer and the supercomputer to allow remote visualization, including interactive remote rendering in the visualization lab using a 3D projector.