Taming the Metal: Platform Engineering for HPC
summary
Alps is a 2,500‑node, 10,000‑GPU system (430 beta flops) turned into immutable, software‑defined clusters using the vCluster framework and a merged ops+dev team; two years of failed separate engineering is replaced by a Git‑driven platform with Terraform, Ansible, Kubernetes, Argo and CI/CD, plus Nomad on bare metal for job scheduling and Terraform provisioning control services. New clusters spawn from pull requests, production reaches 95% availability, delivers millions of GPU hours and ships a 16B cluster in production.
AI-written summary · about
related talks
23mWhen Platform Engineers Lead FinOps: Driving Reliability and $20M in...
24mPlatform Engineering Is the Key Lever for Sustainable Cloud Infrastructure
29mA Multi-Cluster, Multi-Cloud Infrastructure with GitOps at CERN
27mLive Panel: Building at the Roofline — Planning and Delivering Gigawatt-Scale AI Infrastructure
26mTwo sides of the platform coin with operator's and customer's voices
26mMulti-Tenant vs Micro-Clusters in Kubernetes-Driven Platform Engineering
26mGitOps Me Some of That! Managing Hundreds of Clusters with Argo CD
22mIntelligent Autoscaling in Enterprise...