Developers want one thing: GPUs — lots of them — to train models and run inference.
Ordering a cluster from your favorite cloud provider is all fun and games, until data gravity and compliance pull you back on‑prem.
What if you could deliver a cloud‑like developer experience for AI on your own hardware?

In this talk, you’ll hear about our journey (trying) to achieve exactly that. Expect a walk through the architectual and operational challenges of GPU orchestration: Interactions between hypervisors and Kubernetes, scheduling single- and multi-GPU workloads, and trade-offs between virtualized and baremetal clusters.

Come for the lessons learned; stay for the scars and stories from trying to make “GPUs at home” actually work. You’ll leave with tested approaches, lessons learned, and maybe a few warnings.