Serverless Renderfarm

Traditional server farms allow renting individual servers for months. Servers with powerful GPU are very expensive. If you want to run a render server and you are not sure how often and how many jobs to process, a GPU server is not very cost effective. Additionally, a GPU server can only run one rendering job at a time. So the waiting time is extremely high (hours or days) when the number of jobs is high or the benefit does not justify the cost when the number of jobs is low.

Serverless functions are defined tasks that are executed on demand. For this, a virtual server with specified hardware is started up and shut down again after the task. Thus, only the costs of the processing time are incurred and (theoretically) an infinite number of jobs can be started simultaneously. In this project, this approach was implemented and compared with AWS Lambda, Runpod & Argo workflows.