Explore a 35-minute conference talk where Nathan Harper and John Garbutt demonstrate how Graphcore leverages OpenStack and Infrastructure as Code to meet the demanding data requirements of AI systems. Learn about Graphcore's IPU (Intelligence Processing Unit) chips and their integration with high-speed networking interfaces, discovering how the company optimizes infrastructure for maximum data processing efficiency. Understand how the flexible network architecture of IPU-POD products enables scalable processing capabilities, adapting to various workload sizes through resource splitting and joining. Dive into the implementation of modern deployment methods using Terraform, Ansible, and Packer for rapid infrastructure reconfiguration, including critical RDMA via SRIOV functionality. Watch a practical demonstration and walkthrough of Terraform configuration that showcases how Graphcore and StackHPC have moved beyond traditional time-consuming deployment methods to create dynamic, easily reconfigurable platforms and cloud services.
Overview
Syllabus
Satisfying AI's demanding data needs at Graphcore using Infrastructure as Code
Taught by
OpenInfra Foundation