kraze.yml

# NVIDIA GPU cluster example
#
# Prerequisites: nvidia-container-toolkit installed on the host
#   https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html
#
# The default Docker runtime does NOT need to be changed to nvidia.
# GPU passthrough uses Docker's --gpus flag which works with the standard runtime.
#
# Verify setup:
#   nvidia-ctk --version
#   docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi

cluster:
  name: gpu-cluster

  gpu:
    nvidia:
      enabled: true

services:
  nvidia-device-plugin:
    type: helm
    repo: https://nvidia.github.io/k8s-device-plugin
    chart: nvidia-device-plugin
    namespace: nvidia
    version: "0.19.0"
    values_inline: |
      # Remove NFD-based nodeAffinity so the plugin runs without Node Feature Discovery
      affinity: null
      # Mount the kind node's root filesystem so the plugin can find NVIDIA libraries
      nvidiaDriverRoot: "/"
      # Privileged is required to open /dev/nvidiactl for NVML initialization
      securityContext:
        privileged: true