Presented by Yanik Ngoko (PhD, Research Engineer at Qarnot computing) and Clément Pellegrini (Senior Software Engineer at Qarnot computing)
In this presentation, we present a new approach for parallel rendering of images and video. The core computational problem we address is the one decomposing the work such as to ensure a load balance between a set of workers, orchestrated with a master-slave model. In comparison with existing work, we innovate in: (a) introducing a new method for load estimation and (b) proposing a new algorithm for 2D work partitioning. Our proposal is also based on modifications that we propose in the Blender software. The preliminary results we obtain show that the proposed approach can be used to gain several hours in image and video rendering.