Openshift with A100 GPU driver pod not getting ready and nvidia-smi output "No device were found"

[anoopsinghnegi]

1. Quick Debug Information

• OS/Version: Red Hat Enterprise Linux CoreOS release 4.12
• Kernel Version: 4.18.0-372.82.1.el8_6.x86_64
• Container Runtime Type/Version: CRI-O
• K8s Flavor/Version: Openshift Server Version: 4.12.45, Kubernetes Version: v1.25.14+a52e8df
• GPU Operator Version: v23.9.1, v23.6.1
• GPU node: A100

2. Issue or feature description

GPU-operator's driver pod failed to get ready, cluster policy installed default with "use_ocp_driver_toolkit" as selected, driver pod is not getting ready because its Startup probe failed: No devices were found GPU node is of type A100

[root@worker6 driver]# lspci | grep -i nvidia
13:00.0 3D controller: NVIDIA Corporation GA100 [A100 PCIe 40GB] (rev a1)
[root@worker6 driver]#

since the driver pod is not fully ready(container nvidia-driver-ctr is not getting ready) other pods are stuck in init state

[core@master0 tmp]$ kubectl get po -n nvidia-gpu-operator
NAME                                                  READY   STATUS     RESTARTS      AGE
gpu-feature-discovery-mdkg4                           0/1     Init:0/1   0             52m
gpu-operator-595587c664-96gsn                         1/1     Running    0             114m
nvidia-container-toolkit-daemonset-vpmlh              0/1     Init:0/1   0             52m
nvidia-dcgm-2gxdj                                     0/1     Init:0/1   0             52m
nvidia-dcgm-exporter-q96qf                            0/1     Init:0/2   0             52m
nvidia-device-plugin-daemonset-stb74                  0/1     Init:0/1   0             52m
nvidia-driver-daemonset-412.86.202311271639-0-qkdsz   1/2     Running    2 (11m ago)   53m
nvidia-node-status-exporter-fmnkb                     1/1     Running    0             53m
nvidia-operator-validator-h8kwp                       0/1     Init:0/4   0             52m
[core@master0 tmp]$

[core@master0 tmp]$ kubectl exec -ti nvidia-driver-daemonset-412.86.202311271639-0-qkdsz -n nvidia-gpu-operator -- nvidia-smi
No devices were found
command terminated with exit code 6
[core@master0 tmp]$

We have tried with use_ocp_driver_toolkit set to "false" and applied the entitlement but even with it driver pod failed to download packages "kernel-headers-4.18.0-372.82.1.el8_6.x86_64 kernel-devel-4.18.0-372.82.1.el8_6.x86_64" and went into crash-loopback state, the following error is coming

Installing Linux kernel headers...
+ echo 'Installing Linux kernel headers...'
+ dnf -q -y --releasever=8.6 install kernel-headers-4.18.0-372.82.1.el8_6.x86_64 kernel-devel-4.18.0-372.82.1.el8_6.x86_64
Error: Unable to find a match: kernel-headers-4.18.0-372.82.1.el8_6.x86_64 kernel-devel-4.18.0-372.82.1.el8_6.x86_64
++ rm -rf /tmp/tmp.ffQrW5HEcg

3. Steps to reproduce the issue

Use following manifest YML to create cluster-policy

apiVersion: nvidia.com/v1
kind: ClusterPolicy
metadata:
  name: gpu-cluster-policy
spec:
  vgpuDeviceManager:
    config:
      default: default
    enabled: true
  migManager:
    config:
      default: all-disabled
      name: default-mig-parted-config
    enabled: true
  operator:
    defaultRuntime: crio
    initContainer: {}
    runtimeClass: nvidia
    use_ocp_driver_toolkit: true
  dcgm:
    enabled: true
  gfd:
    enabled: true
  dcgmExporter:
    config:
      name: ''
    enabled: true
    serviceMonitor:
      enabled: true
  cdi:
    default: false
    enabled: false
  driver:
    certConfig:
      name: ''
    enabled: true
    kernelModuleConfig:
      name: ''
    licensingConfig:
      configMapName: ''
      nlsEnabled: false
    repoConfig:
      configMapName: ''
    upgradePolicy:
      autoUpgrade: true
      drain:
        deleteEmptyDir: false
        enable: false
        force: false
        timeoutSeconds: 300
      maxParallelUpgrades: 1
      maxUnavailable: 25%
      podDeletion:
        deleteEmptyDir: false
        force: false
        timeoutSeconds: 300
      waitForCompletion:
        timeoutSeconds: 0
    virtualTopology:
      config: ''
  devicePlugin:
    config:
      default: ''
      name: ''
    enabled: true
  kataManager:
    config:
      artifactsDir: /opt/nvidia-gpu-operator/artifacts/runtimeclasses
  mig:
    strategy: single
  sandboxDevicePlugin:
    enabled: true
  validator:
    plugin:
      env:
        - name: WITH_WORKLOAD
          value: 'false'
  nodeStatusExporter:
    enabled: true
  daemonsets:
    rollingUpdate:
      maxUnavailable: '1'
    updateStrategy: RollingUpdate
  sandboxWorkloads:
    defaultWorkload: container
    enabled: false
  gds:
    enabled: false
  vgpuManager:
    enabled: false
  vfioManager:
    enabled: true
  toolkit:
    enabled: true
    installDir: /usr/local/nvidia
status:
  namespace: nvidia-gpu-operator
  state: notReady

let us know if more information is required.

[cdesiniotis]

No devices were found typically indicates the driver failed to initialize. Can you collect system logs by running dmesg | grep -i nvrm on the host?