Use Warewulf to rapidly provision flexible and reproducible AMD AI clusters скачать в хорошем качестве

Use Warewulf to rapidly provision flexible and reproducible AMD AI clusters

At SC25, AMD’s Global Cluster Engineering team shares how they deploy, update, and operate dynamic AI clusters using Warewulf, the open source provisioning system trusted across HPC and AI environments. In this session, Zachary Reid, Michael Hogan, and Massud Zaran walk through real lessons learned from national labs, research facilities, and enterprise datacenters, and explain why Warewulf became the foundation of their rapidly evolving AI infrastructure. If you’re building GPU dense AI clusters or modernizing HPC infrastructure, this session shows how Warewulf eliminates configuration drift, accelerates image and driver updates, simplifies kernel testing, and offers a clean separation of concerns for monitoring and higher level tooling. AMD’s engineers share candid examples of past pitfalls, how Warewulf solved them, and what features they rely on most. What you’ll learn: -Why AI and HPC provisioning challenges are converging, from reproducibility to high speed interconnects to parallel file systems -Operational pitfalls in traditional cluster environments: configuration drift, inconsistent packages, slow reprovisioning, mismatched drivers, poor NUMA configuration, and networking bottlenecks -Why AMD chose Warewulf over tools like xCAT, IPMI workflows, and disk-based provisioning Zachary Reid - Senior Systems Design Engineer, AMD Michael Hogan - Senior HPC Systems Design Engineer, AMD Masood Zaran - Senior Member of Technical Staff and Manager, AMD Resources Warewulf: https://learn.ciq.com/solution-brief/... CIQ: https://ciq.com/ AMD: https://www.amd.com/en.html