Multi-tenant GPUs on
Bare Metal OpenShift AI
A GitOps Blueprint from the Trenches
by Luca Berton · AI & Cloud Advisor
🎬 Watch the Talk
What This Talk Covers
Running GPU workloads on bare metal is hard. Sharing them across teams without chaos is harder. This talk shares a battle-tested framework for building a safe, fair, and efficient multi-tenant GPU platform — everything managed through GitOps.
Safety
Blast radius = zero. RBAC, NetworkPolicy, quotas — all in Git. One team can never break another.
Fairness
Deterministic contention. PriorityClasses, preemption rules, and GPU caps — no more "random wins."
Efficiency
Outcomes per GPU-hour. Time slicing, MIG, GPUDirect RDMA — measured and optimized per tenant.
Key Topics
The Stack
About the Speaker
Luca Berton is an AI & Cloud Advisor with 15+ years building enterprise infrastructure. Author of 8 books including Kubernetes Recipes, creator of the 40,000+ student learning community, and hands-on consultant for GPU platforms, AI integration, and platform engineering.
Based in Amsterdam 🇳🇱 · Previously at Red Hat, Allianz, Databricks
Let's Talk GPU Infrastructure
Building a multi-tenant GPU platform? Migrating to bare metal AI? Considering OpenShift AI?
I'm in Amsterdam for KubeCon — let's grab a coffee and talk shop.
Or find me at the conference — I'm always up for a good conversation about GPU scheduling.
Get insights on AI infrastructure, GPU platforms, and cloud native engineering
📬 Subscribe to the Newsletter