Serverless GPUs in Production: How Cerebrium Built a Globally Efficient Low-Latency AI Platform with Knative

A Maintainer Track session, given with Dave Protasowski, a Knative maintainer.

It covers how Cerebrium, a serverless GPU platform serving customers around the world at low latency and cost, runs on Knative: the operational problems met across different environments, how Cerebrium solved them, and what was learned about running Knative in production for AI workloads.