diff --git a/openshift/install-gpu-ocp.rst b/openshift/install-gpu-ocp.rst index 8b04295f4..bf2fcc592 100644 --- a/openshift/install-gpu-ocp.rst +++ b/openshift/install-gpu-ocp.rst @@ -193,6 +193,11 @@ When you install the **NVIDIA GPU Operator** in the OpenShift Container Platform .. note:: If you create a ClusterPolicy that contains an empty specification such as ``spec{}``, the ClusterPolicy fails to deploy. +You must set the driver version when you create the cluster policy on a newly installed OpenShift Container Platform 5.0 cluster. +Newly installed clusters use RHCOS 10 nodes, and the default driver image does not work on these nodes. +A cluster that you upgraded from OpenShift 4.x keeps RHCOS 9 nodes and is not affected. +Refer to :ref:`openshift-known-issues-ocp5-default-driver`. + As a cluster administrator, you can create a ClusterPolicy using the OpenShift Container Platform CLI or the web console. Also, these steps differ when using **NVIDIA vGPU**. Refer to the appropriate sections below. diff --git a/openshift/known-issues.rst b/openshift/known-issues.rst index 41d733920..e41addf76 100644 --- a/openshift/known-issues.rst +++ b/openshift/known-issues.rst @@ -1,4 +1,4 @@ -.. Date: November 18 2026 +.. Date: November 22 2026 .. Author: nbugrovs .. headings are # * - = @@ -118,3 +118,49 @@ To check an image that fails to mirror, use ``skopeo`` to inspect the image mani The Docker Image Manifest V2, Schema 2 specification defines no uncompressed layer media type, so registries reject it. To report an affected image that this page does not list, add the image reference and the ``oc-mirror`` error output to `NVIDIA/gpu-operator#3166 `_. + +.. _openshift-known-issues-ocp5-default-driver: + +********************************************************************************************* +Default driver container fails on newly installed OpenShift Container Platform 5.0 clusters +********************************************************************************************* + +When you install the GPU Operator with the default cluster policy on a newly installed OpenShift Container Platform 5.0 cluster, the ``nvidia-driver-daemonset`` pods fail on the GPU nodes. + +Newly installed clusters use RHCOS 10 nodes. A cluster that you upgraded from OpenShift 4.x keeps RHCOS 9 nodes and is not affected. + +The GPU Operator bundle pins the default driver container image by digest. The image is based on RHEL 9. The Operator uses a digest-pinned driver image as-is, without per-node operating system tag resolution. The RHEL 9 driver container therefore runs on RHCOS 10 nodes and fails to load kernel modules. + +Specify the driver version in the cluster policy +================================================ + +#. Set ``spec.driver.version`` in the cluster policy to a floating driver version, such as ``${recommended}``. The Operator resolves the operating-system tag for each node and deploys ``${recommended}-rhel10`` on RHCOS 10 nodes: + + .. code-block:: yaml + + driver: + repository: nvcr.io/nvidia + image: driver + version: "${recommended}" + + Alternatively, set the field to the RHEL 10 driver image digest for driver version ``${recommended}``: + + .. code-block:: yaml + + driver: + repository: nvcr.io/nvidia + image: driver + version: "sha256:52b984c00817b08dcf5a0ed50dbba7b9f4d779614bde486a3483a4fb61dea3e5" + +#. Verify that the driver daemon set uses the RHEL 10 image: + + .. code-block:: console + + $ oc -n nvidia-gpu-operator get ds -o custom-columns=NAME:.metadata.name,IMAGE:.spec.template.spec.containers[0].image + + *Partial Output* + + .. code-block:: console + + NAME IMAGE + nvidia-driver-daemonset-10.2.20261015-0 nvcr.io/nvidia/driver:${recommended}-rhel10