Skip to main content
Livepeer orchestrators can serve three compute job categories, each with different requirements and operational expectations.

Transcoding

Video transcoding converts streams between formats, bitrates, and resolutions.
  • GPU: Optional (CPU mode is available, GPU is recommended for throughput)
  • Additional software: None beyond go-livepeer
  • Typical workloads: Live streaming and VOD transcoding

Realtime AI

Realtime AI pipelines process live video through models with strict latency targets.
  • GPU: Required
  • Additional software: Stream handling components such as mediamtx
  • Typical workloads: Live video effects and interactive AI video
Realtime AI setup is more involved than transcoding. Review box.md before planning deployment.

Batch AI

Batch AI handles asynchronous inference jobs without realtime latency constraints.
  • GPU: Required
  • Additional software: Pipeline-dependent runtimes
  • Typical workloads: Offline processing, analysis, and post-production AI
Last modified on March 2, 2026