ROCm · ci-penbot-01 · Apr 6, 2026
diff --git a/Makefile b/Makefile
@@ -411,7 +411,8 @@ helm-k8s: helmify manifests kustomize clean-helm gen-kmm-charts gen-remediation-
 	$(MAKE) helm-docs
 	echo "dependency update, lint and pack charts"
 	cd $(shell pwd)/helm-charts-k8s; helm dependency update; helm lint .; cd ..; helm package helm-charts-k8s/ --destination ./helm-charts-k8s
-	mv $(shell pwd)/helm-charts-k8s/gpu-operator-charts-$(HELM_CHART_VERSION).tgz $(GPU_OPERATOR_CHART)
+	# Avoid $(GPU_OPERATOR_CHART) here: when exported as the chart dir, mv would be a no-op/same-file error.
+	mv $(shell pwd)/helm-charts-k8s/gpu-operator-charts-$(HELM_CHART_VERSION).tgz $(shell pwd)/helm-charts-k8s/$(HELM_OUTPUT_FILE_NAME)
 
 .PHONY: bundle-build
 bundle-build: operator-sdk manifests kustomize ## OpenShift Build OLM bundle.
@@ -602,6 +603,10 @@ helmify:
 helm-install: ## Deploy Helm Charts.
 	helm install -f helm-charts-k8s/values.yaml amd-gpu-operator ${GPU_OPERATOR_CHART} -n kube-amd-gpu --create-namespace ${SKIP_NFD_CMD} ${SKIP_KMM_CMD} ${SKIP_REMEDIATION_CONTROLLER_CMD} ${HELM_OC_CMD} ${SIM_ENABLE_CMD} ${SKIP_INSTALL_DEFAULT_CR_CMD}
 
+.PHONY: helm-upgrade-install
+helm-upgrade-install: ## Same as helm-install but safe when release amd-gpu-operator already exists.
+	helm upgrade --install -f helm-charts-k8s/values.yaml amd-gpu-operator ${GPU_OPERATOR_CHART} -n kube-amd-gpu --create-namespace ${SKIP_NFD_CMD} ${SKIP_KMM_CMD} ${SKIP_REMEDIATION_CONTROLLER_CMD} ${HELM_OC_CMD} ${SIM_ENABLE_CMD} ${SKIP_INSTALL_DEFAULT_CR_CMD}
+
 .PHONY: helm-uninstall
 helm-uninstall-k8s: ## Undeploy Helm Charts.
 	echo "Deleting all device configs before uninstalling operator..."

diff --git a/api/v1alpha1/deviceconfig_types.go b/api/v1alpha1/deviceconfig_types.go
@@ -602,7 +602,9 @@ type ConfigManagerSpec struct {
 	// +optional
 	ImageRegistrySecret *v1.LocalObjectReference `json:"imageRegistrySecret,omitempty"`
 
-	// config map to customize the config for config manager, if not specified default config will be applied
+	// ConfigMap holding DCM config.json. When set, the operator mounts this ConfigMap and does not create it.
+	// When omitted or name is empty, the operator mounts ConfigMap "default-dcm-config" and creates it in the
+	// DeviceConfig namespace if it does not already exist (same default payload as chart defaultDCMConfigMap).
 	//+operator-sdk:csv:customresourcedefinitions:type=spec,displayName="Config",xDescriptors={"urn:alm:descriptor:com.amd.deviceconfigs:configmap"}
 	// +optional
 	Config *v1.LocalObjectReference `json:"config,omitempty"`

diff --git a/bundle/manifests/amd-gpu-operator.clusterserviceversion.yaml b/bundle/manifests/amd-gpu-operator.clusterserviceversion.yaml
@@ -35,8 +35,8 @@ metadata:
       ]
     capabilities: Seamless Upgrades
     categories: AI/Machine Learning,Monitoring
-    containerImage: docker.io/rocm/amd-gpu-operator:dev
-    createdAt: "2026-04-02T12:26:30Z"
+    containerImage: registry.test.pensando.io:5000/amd-gpu-operator:dev
+    createdAt: "2026-04-06T08:31:30Z"
     description: |-
       Operator responsible for deploying AMD GPU kernel drivers, device plugin, device test runner and device metrics exporter
       For more information, visit [documentation](https://instinct.docs.amd.com/projects/gpu-operator/en/latest/)
@@ -127,8 +127,10 @@ spec:
         path: configManager
         x-descriptors:
         - urn:alm:descriptor:com.amd.deviceconfigs:configManager
-      - description: config map to customize the config for config manager, if not
-          specified default config will be applied
+      - description: ConfigMap holding DCM config.json. When set, the operator mounts
+          this ConfigMap and does not create it. When omitted or name is empty, the
+          operator mounts ConfigMap "default-dcm-config" and creates it in the DeviceConfig
+          namespace if it does not already exist (same default payload as chart defaultDCMConfigMap).
         displayName: Config
         path: configManager.config
         x-descriptors:

diff --git a/bundle/manifests/amd.com_deviceconfigs.yaml b/bundle/manifests/amd.com_deviceconfigs.yaml
@@ -111,8 +111,10 @@ spec:
                 description: config manager
                 properties:
                   config:
-                    description: config map to customize the config for config manager,
-                      if not specified default config will be applied
+                    description: |-
+                      ConfigMap holding DCM config.json. When set, the operator mounts this ConfigMap and does not create it.
+                      When omitted or name is empty, the operator mounts ConfigMap "default-dcm-config" and creates it in the
+                      DeviceConfig namespace if it does not already exist (same default payload as chart defaultDCMConfigMap).
                     properties:
                       name:
                         default: ""

diff --git a/config/crd/bases/amd.com_deviceconfigs.yaml b/config/crd/bases/amd.com_deviceconfigs.yaml
@@ -107,8 +107,10 @@ spec:
                 description: config manager
                 properties:
                   config:
-                    description: config map to customize the config for config manager,
-                      if not specified default config will be applied
+                    description: |-
+                      ConfigMap holding DCM config.json. When set, the operator mounts this ConfigMap and does not create it.
+                      When omitted or name is empty, the operator mounts ConfigMap "default-dcm-config" and creates it in the
+                      DeviceConfig namespace if it does not already exist (same default payload as chart defaultDCMConfigMap).
                     properties:
                       name:
                         default: ""

diff --git a/config/manifests/bases/amd-gpu-operator.clusterserviceversion.yaml b/config/manifests/bases/amd-gpu-operator.clusterserviceversion.yaml
@@ -94,8 +94,10 @@ spec:
         path: configManager
         x-descriptors:
         - urn:alm:descriptor:com.amd.deviceconfigs:configManager
-      - description: config map to customize the config for config manager, if not
-          specified default config will be applied
+      - description: ConfigMap holding DCM config.json. When set, the operator mounts
+          this ConfigMap and does not create it. When omitted or name is empty, the
+          operator mounts ConfigMap "default-dcm-config" and creates it in the DeviceConfig
+          namespace if it does not already exist (same default payload as chart defaultDCMConfigMap).
         displayName: Config
         path: configManager.config
         x-descriptors:

diff --git a/docs/dcm/device-config-manager-configmap.md b/docs/dcm/device-config-manager-configmap.md
@@ -2,6 +2,18 @@
 
 The Device Config Manager (DCM) job is to monitor for and apply different configurations on nodes in your cluster. This is done by defining different profiles that can then be applied to each node on your cluster. As such, DCM relies on a Kubernetes ConfigMap that contains the definitions of each configuration profile. This ConfigMap is required to be present for the Device Config Manager to function properly. Once profiles have been defined, specific node labels can be put on the nodes in the cluster to specify which profile should be applied. DCM monitors for any changes in the ConfigMap or changes to the profile node label and applies the correct configuration accordingly. This ConfigMap approach helps to simplify the rollout of different config profiles across all the nodes in the cluster.
 
+## How the GPU Operator uses this ConfigMap
+
+When **Device Config Manager** is enabled on a **DeviceConfig** (`spec.configManager.enable: true`), the GPU Operator deploys DCM on nodes that match the DeviceConfig selector.
+
+**Default ConfigMap name.** If you do **not** set `spec.configManager.config` (or you leave the referenced name empty), the operator mounts a ConfigMap named **`default-dcm-config`** in the **same namespace as the DeviceConfig**. If that object is missing, the operator **creates** it and fills it with a built-in default `config.json` (partition profiles and related settings suitable for typical use). If `default-dcm-config` already exists—for example because your Helm install created it—the operator **does not** overwrite it.
+
+**Custom ConfigMap.** If you set `spec.configManager.config` to a specific ConfigMap name, the operator mounts **that** ConfigMap only. You must create and maintain it; the operator does not populate it for you.
+
+**Where DCM reads config.** The ConfigMap is mounted into the DCM container so profiles are available under **`/etc/config-manager/`**, usually as **`config.json`** in the ConfigMap’s `data`.
+
+**Helm install.** The GPU Operator Helm chart can optionally install the same **`default-dcm-config`** object in the release namespace at deploy time, so the ConfigMap may already be present before you apply a DeviceConfig. Whether it comes from Helm or from the operator, behavior is the same: the default name is used when you omit a custom reference, and an existing object is preserved.
+
 ## ConfigMap
 
 As mentioned, the `config.json` data specifies different GPU partitioning profiles that can be set on the GPU nodes in your cluster. Below is an example Device Config Manager ConfigMap. This example ConfigMap is also available in the GPU Operator repo here: [_example/configmap.yaml_](https://github.com/ROCm/gpu-operator/blob/main/example/configManager/configmap.yaml)

diff --git a/example/configManager/deviceconfigs_example.yaml b/example/configManager/deviceconfigs_example.yaml
@@ -12,7 +12,7 @@ spec:
     enable: True
 
     # image for the device-config-manager container
-    image: "rocm/device-config-manager:v1.4.0"
+    image: "rocm/device-config-manager:v1.4.1"
 
     # image pull policy for config manager set to always to pull image of latest version
     imagePullPolicy: Always

diff --git a/hack/k8s-patch/metadata-patch/values.yaml b/hack/k8s-patch/metadata-patch/values.yaml
@@ -361,6 +361,77 @@ kmmModuleLoader:
   serviceAccount:
     annotations: {}
 kubernetesClusterDomain: cluster.local
+# Default ConfigMap for Device Config Manager when DeviceConfig.spec.configManager.config is omitted.
+# The name must match internal/configmanager.DefaultDCMConfigMapName ("default-dcm-config").
+# Override .data to supply partition profiles; see docs/dcm/device-config-manager-configmap.md.
+defaultDCMConfigMap:
+  # -- Install the default ConfigMap in the operator release namespace (recommended when using DCM without an explicit config reference).
+  install: true
+  # -- ConfigMap metadata.name; must stay aligned with the operator default mount name.
+  name: default-dcm-config
+  # -- Keys become ConfigMap data keys (typically config.json with DCM JSON content).
+  data:
+    config.json: |
+      {
+        "gpu-config-profiles": {
+          "default": {
+            "profiles": [
+              {
+                "computePartition": "SPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          },
+          "cpx_nps1_all": {
+            "profiles": [
+              {
+                "computePartition": "CPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          },
+          "cpx_nps4_all": {
+            "profiles": [
+              {
+                "computePartition": "CPX",
+                "memoryPartition": "NPS4"
+              }
+            ]
+          },
+          "dpx_nps2_all": {
+            "profiles": [
+              {
+                "computePartition": "DPX",
+                "memoryPartition": "NPS2"
+              }
+            ]
+          },
+          "qpx_nps1_all": {
+            "profiles": [
+              {
+                "computePartition": "QPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          },
+          "heterogeneous_example": {
+            "profiles": [
+              {
+                "computePartition": "CPX",
+                "memoryPartition": "NPS1",
+                "numGPUsAssigned": 2
+              },
+              {
+                "computePartition": "SPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          }
+        },
+        "gpuClientSystemdServices": {
+          "names": ["amd-metrics-exporter", "gpuagent"]
+        }
+      }
 managerConfig:
   controllerManagerConfigYaml: |-
     healthProbeBindAddress: :8081

diff --git a/helm-charts-k8s/Chart.lock b/helm-charts-k8s/Chart.lock
@@ -9,4 +9,4 @@ dependencies:
   repository: file://./charts/remediation-crds
   version: v1.0.0
 digest: sha256:0806f6b6d7aa21be77bf1c91e720ae3238338a16f107df450a53b02ef940db1b
-generated: "2026-04-02T12:26:25.920315689Z"
+generated: "2026-04-06T08:31:27.592910404Z"
diff --git a/helm-charts-k8s/README.md b/helm-charts-k8s/README.md
@@ -125,7 +125,7 @@ The AMD GPU Operator is licensed under the [Apache License 2.0](LICENSE).
 
 ## gpu-operator-charts
 
-![Version: v0.0.1](https://img.shields.io/badge/Version-v0.0.1-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: dev](https://img.shields.io/badge/AppVersion-dev-informational?style=flat-square)
+![Version: v0.0.1](https://img.shields.io/badge/Version-v0.0.1-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: dev](https://img.shields.io/badge/AppVersion-test--img-informational?style=flat-square)
 
 AMD GPU Operator simplifies the deployment and management of AMD Instinct GPU accelerators within Kubernetes clusters.
 
@@ -164,6 +164,9 @@ Kubernetes: `>= 1.29.0-0`
 | controllerManager.nodeSelector | object | `{}` | Node selector for AMD GPU operator controller manager deployment |
 | crds.defaultCR.install | bool | `true` | Deploy default DeviceConfig during helm chart installation |
 | crds.defaultCR.upgrade | bool | `false` | Deploy / Patch default DeviceConfig during helm chart upgrade. Be careful about this option: 1. Your customized change on default DeviceConfig may be overwritten 2. Your existing DeviceConfig may conflict with upgraded default DeviceConfig  |
+| defaultDCMConfigMap.data | object | `{"config.json":"{\n  \"gpu-config-profiles\": {\n    \"default\": {\n      \"profiles\": [\n        {\n          \"computePartition\": \"SPX\",\n          \"memoryPartition\": \"NPS1\"\n        }\n      ]\n    },\n    \"cpx_nps1_all\": {\n      \"profiles\": [\n        {\n          \"computePartition\": \"CPX\",\n          \"memoryPartition\": \"NPS1\"\n        }\n      ]\n    },\n    \"cpx_nps4_all\": {\n      \"profiles\": [\n        {\n          \"computePartition\": \"CPX\",\n          \"memoryPartition\": \"NPS4\"\n        }\n      ]\n    },\n    \"dpx_nps2_all\": {\n      \"profiles\": [\n        {\n          \"computePartition\": \"DPX\",\n          \"memoryPartition\": \"NPS2\"\n        }\n      ]\n    },\n    \"qpx_nps1_all\": {\n      \"profiles\": [\n        {\n          \"computePartition\": \"QPX\",\n          \"memoryPartition\": \"NPS1\"\n        }\n      ]\n    },\n    \"heterogeneous_example\": {\n      \"profiles\": [\n        {\n          \"computePartition\": \"CPX\",\n          \"memoryPartition\": \"NPS1\",\n          \"numGPUsAssigned\": 2\n        },\n        {\n          \"computePartition\": \"SPX\",\n          \"memoryPartition\": \"NPS1\"\n        }\n      ]\n    }\n  },\n  \"gpuClientSystemdServices\": {\n    \"names\": [\"amd-metrics-exporter\", \"gpuagent\"]\n  }\n}\n"}` | Keys become ConfigMap data keys (typically config.json with DCM JSON content). |
+| defaultDCMConfigMap.install | bool | `true` | Install the default ConfigMap in the operator release namespace (recommended when using DCM without an explicit config reference). |
+| defaultDCMConfigMap.name | string | `"default-dcm-config"` | ConfigMap metadata.name; must stay aligned with the operator default mount name. |
 | deviceConfig.spec.commonConfig.initContainerImage | string | `"busybox:1.36"` | init container image |
 | deviceConfig.spec.commonConfig.utilsContainer.image | string | `"docker.io/rocm/amd-gpu-operator-utils:latest"` | gpu operator utility container image |
 | deviceConfig.spec.commonConfig.utilsContainer.imagePullPolicy | string | `"IfNotPresent"` | utility container image pull policy |

diff --git a/helm-charts-k8s/crds/deviceconfig-crd.yaml b/helm-charts-k8s/crds/deviceconfig-crd.yaml
@@ -116,8 +116,10 @@ spec:
                 description: config manager
                 properties:
                   config:
-                    description: config map to customize the config for config manager,
-                      if not specified default config will be applied
+                    description: |-
+                      ConfigMap holding DCM config.json. When set, the operator mounts this ConfigMap and does not create it.
+                      When omitted or name is empty, the operator mounts ConfigMap "default-dcm-config" and creates it in the
+                      DeviceConfig namespace if it does not already exist (same default payload as chart defaultDCMConfigMap).
                     properties:
                       name:
                         default: ""

diff --git a/helm-charts-k8s/values.yaml b/helm-charts-k8s/values.yaml
@@ -361,6 +361,77 @@ kmmModuleLoader:
   serviceAccount:
     annotations: {}
 kubernetesClusterDomain: cluster.local
+# Default ConfigMap for Device Config Manager when DeviceConfig.spec.configManager.config is omitted.
+# The name must match internal/configmanager.DefaultDCMConfigMapName ("default-dcm-config").
+# Override .data to supply partition profiles; see docs/dcm/device-config-manager-configmap.md.
+defaultDCMConfigMap:
+  # -- Install the default ConfigMap in the operator release namespace (recommended when using DCM without an explicit config reference).
+  install: true
+  # -- ConfigMap metadata.name; must stay aligned with the operator default mount name.
+  name: default-dcm-config
+  # -- Keys become ConfigMap data keys (typically config.json with DCM JSON content).
+  data:
+    config.json: |
+      {
+        "gpu-config-profiles": {
+          "default": {
+            "profiles": [
+              {
+                "computePartition": "SPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          },
+          "cpx_nps1_all": {
+            "profiles": [
+              {
+                "computePartition": "CPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          },
+          "cpx_nps4_all": {
+            "profiles": [
+              {
+                "computePartition": "CPX",
+                "memoryPartition": "NPS4"
+              }
+            ]
+          },
+          "dpx_nps2_all": {
+            "profiles": [
+              {
+                "computePartition": "DPX",
+                "memoryPartition": "NPS2"
+              }
+            ]
+          },
+          "qpx_nps1_all": {
+            "profiles": [
+              {
+                "computePartition": "QPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          },
+          "heterogeneous_example": {
+            "profiles": [
+              {
+                "computePartition": "CPX",
+                "memoryPartition": "NPS1",
+                "numGPUsAssigned": 2
+              },
+              {
+                "computePartition": "SPX",
+                "memoryPartition": "NPS1"
+              }
+            ]
+          }
+        },
+        "gpuClientSystemdServices": {
+          "names": ["amd-metrics-exporter", "gpuagent"]
+        }
+      }
 managerConfig:
   controllerManagerConfigYaml: |-
     healthProbeBindAddress: :8081