Automating GPU Infrastructure for Kubernetes
summary
Shows an end-to-end solution that containerizes the NVIDIA installer to auto-build and install kernel modules on Container Linux, running the build as a daemonset that pulls the matching kernel sources and toolchain, modprobes the modules and creates device nodes, then uses the Kubernetes device plugin to expose GPUs. Demo runs a darknet object‑detection container built on NVIDIA CUDA and reports GPUs as healthy; it works but still has per-node rebuilds, OS‑upgrade downtime and multiple OS‑specific installers causing fragmentation.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:10Disclaimer
- 1:34About me
- 2:15Collaborating
- 4:15Pop Quiz
- 4:54Example CUDA workload
- 5:25CUDA libraries
- 6:03Linux fragmentation
- 8:54Automating kernel modules
- 9:48CoreOS Nvidia
- 15:36Device Plugins
- 16:32Google Device Plugin
- 18:21Demo
- 24:26Conclusions
related talks
39m✨ GPU in My Containers!
30mExplain How Kubernetes Works With GPU Like I’m 5
30mRunning AI Workloads in Containers and Kubernetes
36mProduction GPU Cluster with K8s for AI and DL Workloads
32mGPU Machine Learning From Laptop to Cloud
26mState of the GPU(nion)
41mBuilding & Operating a Large-scale HPC AI Cluster on Kubernetes
37mBuilding Distributed TensorFlow Using Both GPU and CPU on Kubernetes [I]