First Principles: Superclusters with RDMA—Ultra-high Performance at Massive Scale

Find out more: http://www.oracle.com/goto/fpsuperclu

This First Principles video complements our previous episode: First Principles: building a high performance network in the public cloud. In this episode Pradeep Vincent and Jag Brar, Oracle Cloud Infrastructure architects, explain how OCI took its application of RDMA networking a step further, building superclusters with the help of NVIDIA's ConnectX RDMA NICs to support tens of thousands of GPUs.

00:00 Introduction to RDMA
02:01 What are superclusters with RDMA?
04:11 RDMA superclusters network fabric
04:54 What are the superclusters latency between GPUs?
07:23 Handling latency sensitive workloads in superclusters
08:39 Minimizing latency for GPU workloads in superclusters

For more architectural breakdowns, catch up on more First Principles videos:
First Principles: using redundancy and recovery to achieve high durability in OCI Object Storage:    • First Principles: using redundancy and rec...  
First Principles: making Kubernetes serverless with OCI's Virtual Nodes:    • First Principles: making Kubernetes server...  
First Principles: building a high performance network in the public cloud:    • First Principles: building a high performa...  
First Principles: inside OCI Container Instances:    • First Principles: inside OCI Container Ins...  

Contact Cloud Platform Sales: https://social.ora.cl/6004zuD5G
Oracle Cloud Infrastructure (OCI) - https://www.oracle.com/cloud/
Cloud Storage - https://www.oracle.com/cloud/storage/
Cloud Compute - https://www.oracle.com/cloud/compute/
Cloud Networking - https://www.oracle.com/cloud/networking/
Multicloud - https://www.oracle.com/cloud/multicloud/
Cloud Economics - https://www.oracle.com/cloud/economics/
Cloud Computing Defined - https://www.oracle.com/cloud/what-is-...
Subscribe to Oracle's YouTube channel - https://social.ora.cl/6004yGDBE