Ray-Trace rendering, like MoonRay, has been industry standard for years because of the visual quality and life-like images it can generate. But Ray trace rendering is very compute intensive as it simulates the physics of how light travels and reflects in the real world - millions of rays. Moonray was designed to be uniquely scalable with almost linear scalability to thousands of cores. The challenge is how to use cores beyond the confines of a single compute node and our solution was Arras.
Arras is a scalable computational framework for composing distributed computations into cohesive assemblies capable of complex data manipulation (Engines) such as MCRT rendering, model deformation, physical simulation, etc. Engines like MoonRay can be split across multiple physical machines, allowing computations that are traditionally bound by the resource limitations of a single physical host to leverage potentially unlimited computational power to accelerate processor-intensive operations.
This talk will give an overview of how Arras works and how it is used to leverage MoonRays scalability as well as the various workflows it has enabled in the production of DreamWorks Animation’s films such as Everest, Trolls World Tour, Puss In Boots: The Last Wish and many more.
Session author's bio
Randy is sr. manager of rendering, shading and machine learning teams at DreamWorks, having started in 2015 as his first foray into the animation and the film industry. Randy's led the efforts of advancing MoonRay and it's feature set and related technologies since prior to it's usage on the first feature, How To Train Your Dragon: The Hidden World in 2017, through open-sourcing it in early 2023. He enjoys tool-building and seeing artists push the capabilities beyond expectations.
|Level of Difficulty||Beginner|