Render Farm
A render farm is a cluster of networked computers used to render large quantities of 3D animation frames in parallel, allowing a studio to produce in days what would take a single workstation months.
A single complex 3D frame can take from minutes to hours to render depending on resolution, lighting setup, and effects. A 60-second sequence at 24 fps is 1,440 frames. Rendering linearly on one machine could take weeks; distributing across 50 machines compresses the same job into hours.
Modern render farms split between on-premise hardware (machines a studio owns and operates) and cloud rendering (services like AWS Thinkbox Deadline, GCP, or specialist providers like ZYNC and Conductor). Cloud is more flexible and scales to peak demand; on-premise is more cost-effective for sustained baseline load. Many studios use both.
On hybrid AI projects like LEGS, the render farm handles the 3D side of the pipeline. AI-generated frames come from a different compute pool (GPU-heavy cloud), but the broadcast deliverable is composited at the render farm with both streams converging at the finishing pass.
Render farm cost is now a meaningful part of project budget on heavy 3D work, alongside artist time. We model the render budget at the start of pre-production so it does not surprise anyone at finishing.
Related
Related concepts
Related services
Sources
Academic papers, recognised industry standards, and canonical industry texts that back up claims in this entry.
- Physically Based Rendering: From Theory to Implementation. Pharr, M., Jakob, W., Humphreys, G., MIT Press, 2023Supports: Canonical reference on physically based rendering at production scale
- Production Rendering: Design and Implementation. Apodaca, A. A., Gritz, L., Springer, 2002Supports: Industry-standard text on production rendering and farm scheduling
- Computer Animation: Algorithms and Techniques. Parent, R., Morgan Kaufmann (Elsevier), 2012Supports: Standard textbook on rendering pipelines used in render farms
Frequently asked questions
Do you have your own render farm?
We use a combination of in-house workstations and cloud rendering services depending on project load. For most projects, cloud is the right answer because it scales with the schedule. For sustained workloads, on-premise hardware is more cost-effective.
How long does a typical project render?
Highly variable. A 60-second 3D piece with moderate complexity might take a few hundred GPU-hours; a heavier piece with simulation, complex lighting, and effects can take ten times that. We estimate the render budget per project based on the look development target.
Does AI affect render times?
It can shorten them in some cases (AI-driven upscaling lets us render at lower resolution and up-rate at finishing). It lengthens them in others (AI-generated frames usually need a clean-up pass that is itself compute-intensive). The net effect is project-specific.