Building Custom GPU Clusters at Scale: Using Kubespray To Create High-Performance AI Infrastructure
Talkat KubeCon + CloudNativeCon China 2025 · on Wednesday, 2025-06-11· at 14:30· Level 21 | Pearl Pavilion·
summary
Explains how to use Kubespray to deploy production-ready bare-metal Kubernetes GPU clusters for AI workloads, covering Kubespray fundamentals, features, and recent updates. Describes practical experience and best practices for scaling custom GPU clusters and integrating Kubernetes technologies such as LWS, Kueue, Gateway API Inference Extension, DRA, tensor parallelism to optimize RAG and LoRA workloads, and provides Kubespray inventory source code plus guidance on using Kubernetes operators to define and scale infrastructure in private clouds.
AI-written summary · about
related talks
12mKubespray: Driving Cost-Efficiency for AI on Kubernetes
36mProduction GPU Cluster with K8s for AI and DL Workloads
41mBuilding & Operating a Large-scale HPC AI Cluster on Kubernetes
16mAI + Kubernetes: What Beginners...
34mDeep Dive: Kubernetes/Kubespray (Cluster Lifecycle) SIG
25mExtending Kubernetes for AI | Lessons Learned From Platform Engineering
32mDeep Dive: Kubespray (a sig-cluster-lifecycle's project)
32mDeep Dive: Kubespray (a sig-cluster-lifecycle's project)