Scaling Up Your Cloud with Oracle Superclusters
summary
Oracle describes a cloud design that assembles 131,072 GPUs into superclusters, claims up to 104 petabits per second GPU-to-GPU networking, uses liquid cooling with NVIDIA partnership, and exposes bare metal GPUs plus high-performance storage and RDMA networking for training massive models faster. Oracle says orchestration, observability and rapid capacity build‑out (10,000 GPUs every 10 days) solve repair, isolation and sovereignty needs for customers running large training and inference workloads.
AI-written summary · about
chapters (11)
- 0:00Introduction to scaling
- 1:05Defining super clusters
- 3:04Engineering challenges
- 5:57Need for high performance
- 8:09Networking secret sauce
- 9:30Oracle differentiation
- 12:38Security by design
- 14:59Sovereignty and deployment
- 17:07Customer success stories
- 18:37Zero trust networking
- 22:19Future of OCI
related talks
11mthe future of cloud
01h 13mEngineering the Cloud for Tomorrow: Performance, Speed and Scale for AI Workloads
15mLeo Leung, Oracle | Nvidia GTC 2024
01h 19mAccelerating Growth in the Cloud
01h 07mCloud Generation 2
13mOptimizing Oracle Database for Multicloud
07mMerging big data and cloud
27mFrom Supercomputing to Serving: A Case Study Delivering Cloud Native Foundation Models