Keynote: Building the Infrastructure that Powers the Future of AI
summary
OpenAI runs research on Kubernetes across Azure AWS and private hardware and scales bursts to 4,000 nodes while moving from single‑pod prototypes to runs that jump from one core to 10,000 cores. Built components include an autoscaler for batch jobs that inspects pod scheduling gaps and provisions exact machines (spot or on demand), a library to deploy distributed TensorFlow jobs, and GPU scheduling and CPU affinity scripts; these tools glue together to run the Universe platform of 1,000+ games and it works in production.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:23What Is Open Ai
- 1:09What Makes Building Infrastructure for a Research Lab Running Deep Learning Different from Infrastructure at a Start-Up
- 5:18Primary Service Layer
- 12:43Why Are We Using Coop
related talks
09mKeynote: Kubernetes for GenAI: The New Operating System of the AI-First World
10mKeynote: Enabling AI with Cloud Native
06mSponsored Keynote: Building Your AI Factory: Wh...
28mScaling AI Infrastructure
12mKeynote: AI Infrastructure- 10mKeynote: Enabling AI with Cloud Native
27mHarnessing Kubernetes for Scalable AI/ML Workloads
11mThe Future of Cloud Native and AI