nccurry · Oct 22, 2026 · Oct 22, 2026 · Oct 22, 2026 · Oct 22, 2026 · Oct 22, 2026
diff --git a/gpu-operator/getting-started.rst b/gpu-operator/getting-started.rst
@@ -421,8 +421,10 @@ To view all the options, run ``helm show values nvidia/gpu-operator``.
      - ``{}``
 
    * - ``dcgmExporter.enabled``
-     - By default, the Operator gathers GPU telemetry in Kubernetes using `DCGM Exporter <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/latest/dcgm-exporter.html>`_. 
+     - By default, the Operator gathers GPU telemetry in Kubernetes using `DCGM Exporter <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/latest/dcgm-exporter.html>`_.
        Set this value to ``false`` to disable it.
+       For all DCGM Exporter values, such as custom metrics files and the ``ServiceMonitor``, refer to
+       `DCGM Exporter Helm Values <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/latest/dcgm-exporter.html#dcgm-exporter-helm-values>`_.
        Available values are ``true`` (default) or ``false``.
      - ``true``
 

diff --git a/gpu-operator/gpu-sharing.rst b/gpu-operator/gpu-sharing.rst
@@ -84,11 +84,38 @@ the mixed MIG strategy.
 Limitations
 ===========
 
-- DCGM-Exporter does not support associating metrics to containers when GPU time-slicing is enabled with the NVIDIA Kubernetes Device Plugin.
 - The Operator does not monitor changes to a time-slicing config map.
   Refer to :ref:`time-slicing-update-config-map`.
 
 
+.. _time-slicing-gpu-metrics:
+
+GPU Metrics with Time-Slicing
+=============================
+
+By default, DCGM Exporter does not associate GPU metrics with the pods that share a GPU through time-slicing or MPS.
+To associate them, set the ``KUBERNETES_VIRTUAL_GPUS`` environment variable to ``true`` in ``dcgmExporter.env``.
+For example, in the Helm values:
+
+.. code-block:: yaml
+
+   dcgmExporter:
+     env:
+     - name: KUBERNETES_VIRTUAL_GPUS
+       value: "true"
+
+Or patch the cluster policy.
+The patch replaces the whole ``env`` list, so include any variables that you already set:
+
+.. code-block:: console
+
+   $ kubectl patch clusterpolicies.nvidia.com/cluster-policy --type=merge \
+       -p '{"spec": {"dcgmExporter": {"env": [{"name": "KUBERNETES_VIRTUAL_GPUS", "value": "true"}]}}}'
+
+For the other DCGM Exporter settings in the Operator, refer to
+`DCGM Exporter with the NVIDIA GPU Operator <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/latest/dcgm-exporter.html>`__.
+
+
 Changes to Node Labels
 ======================
 

diff --git a/gpu-operator/life-cycle-policy.rst b/gpu-operator/life-cycle-policy.rst
@@ -84,6 +84,10 @@ Refer to :ref:`Upgrading the NVIDIA GPU Operator` for more information.
 
 **D** = Default driver, **R** = Recommended driver
 
+.. TODO: When the next GPU Operator release is published, add a column for it.
+   Set DCGM Exporter to v4.7.0-4.9.0 and DCGM to 4.7.0-2 in the new column, or to the versions that the release ships.
+   DCGM Exporter 4.9.0 needs DCGM 4.7 or later, so the two rows must change together.
+
 .. flat-table::
    :header-rows: 2
 

diff --git a/gpu-operator/release-notes.rst b/gpu-operator/release-notes.rst
@@ -33,6 +33,78 @@ Refer to the :ref:`GPU Operator Component Matrix` for a list of software compone
 
 ----
 
+.. DRAFT: Release notes for the next GPU Operator release, which ships DCGM Exporter 4.9.0.
+   Release notes are written when a release is published. Move this text into a new
+   version section above and replace the placeholders, then delete this comment.
+   This is a comment only because the GPU Operator version number and date are not known yet.
+
+   New Features
+   ------------
+
+   * Updated software component versions:
+
+     - NVIDIA DCGM Exporter v4.7.0-4.9.0
+     - NVIDIA DCGM 4.7.0-2
+
+   * DCGM Exporter 4.9.0 renames its default metrics to the DCGM 4.7 field names and changes several units.
+     For example, ``DCGM_FI_DEV_GPU_UTIL`` in percent is now ``DCGM_FI_DEV_GPU_UTIL_RATIO``, a ratio from 0 to 1,
+     and the ``DCGM_FI_DRIVER_VERSION`` label is now ``DCGM_FI_SYSTEM_DRIVER_VERSION``.
+     Before you upgrade, update dashboards, alerts, recording rules, and autoscaler rules, such as HPA or KEDA, that use DCGM Exporter metrics,
+     and rescale thresholds for the unit changes, for example ``> 80`` becomes ``> 0.8`` for GPU utilization.
+     Re-import the Grafana dashboard from the DCGM Exporter repository.
+     For the full list of names and units, refer to the
+     `DCGM Exporter release notes <https://docs.nvidia.com/datacenter/dcgm/latest/release-notes/dcgm-exporter.html>`__.
+
+   Improvements
+   ------------
+
+   * The Operator sets ``/etc/dcgm-exporter/default-counters.csv`` as the default DCGM Exporter metrics file.
+     (`GPU Operator PR #3110 <https://github.com/NVIDIA/gpu-operator/pull/3110>`__)
+
+   * On Red Hat OpenShift Container Platform, the ``accelerator_*`` recording rules read the new DCGM Exporter metric names and fall back to the old names,
+     so the OpenShift console dashboards keep working during the upgrade.
+     (`GPU Operator PR #3110 <https://github.com/NVIDIA/gpu-operator/pull/3110>`__)
+
+   * The DCGM Exporter daemon set has a startup probe that allows up to five minutes for the exporter to start before the liveness probe applies.
+     (`GPU Operator PR #3110 <https://github.com/NVIDIA/gpu-operator/pull/3110>`__)
+
+   * On Red Hat OpenShift Container Platform, ``accelerator_memory_used_bytes`` reports framebuffer memory used in bytes instead of memory-copy utilization,
+     and ``accelerator_memory_total_bytes`` is populated with the default metrics.
+     (`GPU Operator PR #3110 <https://github.com/NVIDIA/gpu-operator/pull/3110>`__)
+
+   * DCGM Exporter 4.9.0 stops at startup on more invalid settings.
+     An invalid value in ``dcgmExporter.env`` or ``dcgmExporter.args`` stops the pod with an error message in the pod log.
+
+   Known Issues
+   ------------
+
+   * DCGM Exporter 4.9.0 needs a DCGM 4.7 or later host engine. With an older host engine, such as DCGM 4.6.1, the exporter exits at startup
+     with the error ``the DCGM hostengine refused these fields: ...``, which lists the fields the host engine does not support.
+     This affects clusters where the Operator runs the standalone DCGM host engine, that is ``dcgm.enabled=true``,
+     which is the default on OpenShift, or a ``GPUCluster`` with standalone DCGM:
+
+     - Upgrade the ``dcgm`` and ``dcgmExporter`` images together.
+       If you set ``dcgm.version`` or ``dcgmExporter.version``, change both.
+     - During the upgrade, the exporter pod on a node can restart a few times until the DCGM pod on that node runs 4.7.
+     - If you run DCGM on the host, upgrade it to 4.7 or later.
+     - If you must keep an older host engine, use a custom metrics file with the older field names.
+       The DCGM Exporter release notes list the old and new names.
+       This pairing is untested and unsupported.
+
+   * If you roll the Operator back to a release whose ``accelerator_*`` rules read only the old metric names,
+     also roll back DCGM Exporter, or the ``accelerator_*`` series are empty.
+     Roll back the ``dcgm`` and ``dcgmExporter`` images together, to the versions that Operator release ships.
+
+   Removals and Deprecations
+   -------------------------
+
+   * ``/etc/dcgm-exporter/dcp-metrics-included.csv`` is deprecated. It has the same metrics as ``default-counters.csv``.
+
+   * The older DCGM field names in custom metrics files still work but are deprecated. NVIDIA recommends the DCGM 4.7 names.
+     Refer to `Custom Metrics Files <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/latest/dcgm-exporter.html#dcgm-exporter-custom-metrics>`__.
+
+   * The ``DCGM_EXPORTER_CONFIGMAP_DATA`` environment variable is deprecated. Use ``dcgmExporter.config.name`` instead.
+
 .. _v26.7.2:
 
 26.7.2

diff --git a/gpu-telemetry/about-telemetry.rst b/gpu-telemetry/about-telemetry.rst
@@ -20,39 +20,27 @@
 About GPU Telemetry
 ###################
 
-Monitoring stacks usually consist of a collector, a time-series database to store metrics and a visualization layer.
-A popular open-source stack is `Prometheus <https://prometheus.io/>`_ used along with `Grafana <https://grafana.com/>`_ as
-the visualization tool to create rich dashboards. Prometheus also includes an `Alertmanager <https://github.com/prometheus/alertmanager>`_,
-to create and manage alerts. Prometheus is deployed along with `kube-state-metrics <https://github.com/kubernetes/kube-state-metrics>`_ and
-`node_exporter <https://github.com/prometheus/node_exporter>`_ to expose cluster-level metrics for Kubernetes API objects and node-level
-metrics such as CPU utilization.
+The NVIDIA GPU Operator deploys NVIDIA DCGM Exporter on every GPU node by default.
+DCGM Exporter reads GPU metrics from `NVIDIA DCGM <https://developer.nvidia.com/dcgm>`__
+and exposes them for `Prometheus <https://prometheus.io/>`__.
+You can view the metrics in Grafana or, on Red Hat OpenShift, in the web console.
 
+This documentation covers what the GPU Operator does for GPU telemetry:
 
-To gather GPU telemetry in Kubernetes, its recommended to use DCGM Exporter.  DCGM Exporter, based on `DCGM <https://developer.nvidia.com/dcgm>`_ exposes
-GPU metrics for Prometheus and can be visualized using Grafana.  DCGM Exporter is architected to take advantage of
-``KubeletPodResources`` `API <https://kubernetes.io/docs/reference/command-line-tools-reference/feature-gates/>`_ and exposes GPU metrics in a format that can be
-scraped by Prometheus. A ``ServiceMonitor`` is also included to expose endpoints.
+* :doc:`dcgm-exporter` describes the resources that the Operator creates,
+  the Operator values for DCGM Exporter, custom metrics files, Prometheus scraping, and a separate DCGM host engine.
+* :external+ocp:doc:`enable-gpu-monitoring-dashboard` describes the GPU dashboard in the OpenShift web console.
 
-.. note::
-   DCGM and DCGM Exporter are deployed by default with the `NVIDIA GPU Operator <https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/>`_.
+For DCGM Exporter itself, such as standalone and Helm installs, the configuration file, command-line options,
+and the list of metrics, refer to the
+`DCGM Exporter documentation <https://docs.nvidia.com/datacenter/dcgm/latest/installation/install-dcgm-exporter.html>`__.
 
-
-*************************
-Benefits of GPU Telemetry
-*************************
-
-Understanding GPU usage provides important insights for IT administrators managing a data center.
-Trends in GPU metrics correlate with workload behavior and make it possible to optimize resource allocation,
-diagnose anomalies, and increase overall data center efficiency. 
-As GPUs become more mainstream, users would like to get access to GPU metrics to monitor GPU resources, just
-like they do today for CPUs.
-
-*************************
+***********
 Quick Links
-*************************
+***********
 
-* `DCGM Exporter GitHub repository <https://github.com/NVIDIA/dcgm-exporter>`_
-* `DCGM Documentation <https://docs.nvidia.com/datacenter/dcgm/latest/>`_
-* `DCGM Exporter Documentation <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/dcgm-exporter.html>`_
-* `Integrating GPU Telemetry into Kubernetes <https://docs.nvidia.com/datacenter/cloud-native/gpu-telemetry/integrating-telemetry-kubernetes.html>`_
-* `NVIDIA GPU Operator Documentation <https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/>`_
+* `NVIDIA GPU Operator documentation <https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/>`__
+* `DCGM Exporter documentation <https://docs.nvidia.com/datacenter/dcgm/latest/installation/install-dcgm-exporter.html>`__
+* `DCGM Exporter release notes <https://docs.nvidia.com/datacenter/dcgm/latest/release-notes/dcgm-exporter.html>`__
+* `DCGM Exporter GitHub repository <https://github.com/NVIDIA/dcgm-exporter>`__
+* `DCGM documentation <https://docs.nvidia.com/datacenter/dcgm/latest/>`__