Skip to content

dynamofl/gpu-kind-setup

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 

Repository files navigation

GPU kind bootstrap

This directory contains a self‑contained helper script for spinning up a GPU-enabled kind cluster using NVIDIA's nvkind.

Contents

File Purpose
bootstrap.sh Main automation script (cluster create → device plugin → smoke test).
nvkind-ingress.yaml kind cluster configuration template consumed by nvkind.

Prerequisites

Install the dependencies first (Docker, kind, kubectl, Helm, Go, Make, jq, nvkind). The root bootstrap/install-deps.sh in this repo can do this for you.

Your host must expose NVIDIA GPUs and have the driver + CUDA toolkit stack installed. If Docker is not already configured for the NVIDIA runtime, run the script with CONFIGURE_TOOLKIT=true.

Usage

# optional: tweak defaults
export CLUSTER_NAME=kind-gpu
export CONFIGURE_TOOLKIT=true      # only needed the first time on a new host

bootstrap/bootstrap.sh

The script will:

  1. Clone/build nvkind (or reuse an existing checkout)
  2. Recreate the named cluster via nvkind using the provided template
  3. Install the NVIDIA device plugin with runtimeClassName=nvidia
  4. Verify that GPU allocatable resources are visible
  5. Run a simple nvidia-smi smoke test pod (skip via SKIP_SMOKE_TEST=true)

Key environment variables:

Variable Default Description
CLUSTER_NAME kind-gpu Name of the kind cluster
NVKIND_DIR ../nvkind Location of the nvkind checkout/binary
TEMPLATE_PATH nvkind-ingress.yaml Cluster template passed to nvkind
DELETE_EXISTING true Whether to delete an existing cluster with the same name
CONFIGURE_TOOLKIT false Configure NVIDIA container toolkit + restart Docker
SKIP_SMOKE_TEST false Skip the CUDA smoke test pod

Once the script prints Cluster '<name>' is ready for GPU workloads, you can start installing your application stack.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages