Intel GPU Base Operator v0.3.0

Release 0.3.0

Minor release for Intel GPU Base Operator.

Please note that while the operator continues to develop, the ClusterPolicy CRD will receive changes, sometimes breaking ones. Helm does not update the CRDs automatically, thus it's left for the cluster admin to update them. Easiest way to achieve this is to remove the previous chart, remove the CRDs and then install the new version. Eventually the CRD will stabilize and CRD update needs become scarce.

To delete CRDs:
kubectl delete crds clusterpolicies.intel.com gpufirmwareupdates.intel.com

Breaking CRD changes:

• ClusterPolicy: DRA's deviceTaint variable was removed due to being invalid.

Changes from 0.2.1 release:

• Support for OpenShift deployment
• Updated Intel dependencies
  • Intel GPU DRA driver 0.11.0
    • Support for KubeVirt
      • gpu-vfio.intel.com device class
      • Manage device binding (manageBinding=true)
  • Intel GPU Plugin 0.36.0
    • Single monitoring resource (gpu.intel.com/monitoring)
    • Device Health state retrieved from XPUMD
    • GPU Levelzero sidecar container dropped
  • Intel XPUMD 2.0.0
• Kueue 0.18.1
  • Improved DRA handling (CEL expressions)
• Helm chart creates namespace automatically
  • Creation can be skipped with "createNamespace=false"
• Various Golang dependency updates
  • Controller-runtime updated from 0.22.4 to 0.24.1

Fixes:

• #17
• #16

Known issues:

• Kueue queues are not update properly when clusterpolicy is patched/edited
• Health handling between device plugin and DRA is inconsistent (DP detects warning severity as Unhealthy, DRA detects critical severity as Unhealthy)

Basic install with Intel GPU DRA Driver:

helm install --create-namespace --namespace "intel-gpu-base-operator" --version 0.3.0 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-base-operator" --version 0.3.0 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dra

Basic install with Intel GPU Device Plugin:

helm install --create-namespace --namespace "intel-gpu-base-operator" --version 0.3.0 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-base-operator" --version 0.3.0 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dp

Collateral images:

ghcr.io/intel/intel-resource-drivers-for-kubernetes/intel-gpu-resource-driver:v0.11.0@sha256:49f38fcbee4f98d748b537b1a728a12f39e56fd208c86dc3fa32ab2162c21197
ghcr.io/intel/xpumanager/xpumd:v2.0.0@sha256:bd6bad4f95274832ce6863c6f7744fc73a610903617d9def2932588138be0b85
docker.io/intel/intel-gpu-plugin:0.36.0@sha256:2db679be62b52ac985169084ca711cab6e6c59fe543ab2ddee58163d6f8d29e0

Intel GPU Base Operator v0.3.0-rc1

Release 0.3.0-rc1

Minor release candidate for Intel GPU Base Operator.

Changes between 0.3.0-rc0 release:

• CRD/clusterpolicy: remove DRA's deviceTaint boolean
• xpumd: use critical severity for temperature limits
  • fix ecc_* states causing warnings
• helm: fixes to clusterrole & clusterrolebinding names (to avoid collisions)
  • remove unnecessary clusterroles
• e2e: check helm uninstall outcome

Known issues:

• Kueue queues are not update properly when clusterpolicy is patched/edited
• Health handling between device plugin and DRA is inconsistent (DP detects warning severity as Unhealthy, DRA detects critical severity as Unhealthy)
  • Partly addressed by xpumd config changes

Basic install with Intel GPU DRA Driver:

helm install --create-namespace --namespace "intel-gpu-base-operator" --version 0.3.0-rc1 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-base-operator" --version 0.3.0-rc1 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dra

Basic install with Intel GPU Device Plugin:

helm install --create-namespace --namespace "intel-gpu-base-operator" --version 0.3.0-rc1 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-base-operator" --version 0.3.0-rc1 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dp

Collateral images:

ghcr.io/intel/intel-resource-drivers-for-kubernetes/intel-gpu-resource-driver:v0.11.0@sha256:49f38fcbee4f98d748b537b1a728a12f39e56fd208c86dc3fa32ab2162c21197
ghcr.io/intel/xpumanager/xpumd:v2.0.0@sha256:bd6bad4f95274832ce6863c6f7744fc73a610903617d9def2932588138be0b85
docker.io/intel/intel-gpu-plugin:0.36.0@sha256:2db679be62b52ac985169084ca711cab6e6c59fe543ab2ddee58163d6f8d29e0

Full Changelog: v0.3.0-rc0...v0.3.0-rc1

Intel GPU Base Operator v0.3.0-rc0

Release 0.3.0-rc0

Minor release candidate for Intel GPU Base Operator.

Changes from 0.2.1 release:

• Support for OpenShift deployment
• Updated Intel dependencies
  • Intel GPU DRA driver 0.11.0
    • Support for KubeVirt
      • gpu-vfio.intel.com device class
      • Manage device binding (manageBinding=true)
  • Intel GPU Plugin 0.36.0
    • Single monitoring resource (gpu.intel.com/monitoring)
    • Device Health state retrieved from XPUMD
    • GPU Levelzero sidecar container dropped
  • Intel XPUMD 2.0.0
• Kueue 0.18.1
  • Improved DRA handling (CEL expressions)
• Helm chart creates namespace automatically
  • Creation can be skipped with "createNamespace=false"
• Various Golang dependency updates
  • Controller-runtime updated from 0.22.4 to 0.24.1

Fixes:

• #17
• #16

Known issues:

• Kueue queues are not update properly when clusterpolicy is patched/edited
• Health handling between device plugin and DRA is inconsistent (DP detects warning severity as Unhealthy, DRA detects critical severity as Unhealthy)

Basic install with Intel GPU DRA Driver:

helm install --create-namespace --namespace "intel-gpu-base-operator" --version 0.3.0-rc0 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-base-operator" --version 0.3.0-rc0 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dra

Basic install with Intel GPU Device Plugin:

helm install --create-namespace --namespace "intel-gpu-base-operator" --version 0.3.0-rc0 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-base-operator" --version 0.3.0-rc0 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dp

Collateral images:

ghcr.io/intel/intel-resource-drivers-for-kubernetes/intel-gpu-resource-driver:v0.11.0@sha256:49f38fcbee4f98d748b537b1a728a12f39e56fd208c86dc3fa32ab2162c21197
ghcr.io/intel/xpumanager/xpumd:v2.0.0@sha256:bd6bad4f95274832ce6863c6f7744fc73a610903617d9def2932588138be0b85
docker.io/intel/intel-gpu-plugin:0.36.0@sha256:2db679be62b52ac985169084ca711cab6e6c59fe543ab2ddee58163d6f8d29e0

Intel GPU Base Operator v0.2.1

Release 0.2.1.

First public release for Intel GPU Base Operator.

Functionality:

• Deploy either Intel GPU DRA driver or Intel GPU device plugin
• GPU monitoring via Intel XPUManager Daemon (xpumd)
• Scheduling aided with Node Feature Discovery
• Sets up metrics flow to Prometheus
• Allows creating Kueue queues for advanced scheduling
• Initial support for GPU FW update

Known issues:

• Kueue queues are not update properly when clusterpolicy is patched/edited
• DRA resource requests with cel expressions are not supported by Kueue queues

Basic install with Intel GPU DRA Driver:

kubectl create ns intel-gpu-operator
kubectl label ns intel-gpu-operator resource.kubernetes.io/admin-access=true

helm install --namespace "intel-gpu-operator" --version 0.2.1 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-operator" --version 0.2.1 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dra

Basic install with Intel GPU Device Plugin:

helm install --namespace "intel-gpu-operator" --version 0.2.1 gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart --wait
helm install --namespace "intel-gpu-operator" --version 0.2.1 gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dp

Collateral images:

ghcr.io/intel/intel-resource-drivers-for-kubernetes/intel-gpu-resource-driver:v0.10.0@sha256:746150e64010881dbfdaeb74771703b13cac365a89ee47c4d7499d686ea4163f
ghcr.io/intel/xpumanager/xpumd:v2.0.0-rc.0@sha256:8f020012f68888314402c0332a53718ace4ade9913476bbd125af89edb760a8b
docker.io/intel/intel-gpu-plugin:0.35.0@sha256:34697f9c286857da986381595ac2a693524a83c831955247dae47dfda4d2f526
docker.io/intel/intel-gpu-levelzero:0.35.0@sha256:a8a0729f253de6e8e117a7ef621883a1228f7304747076dcd1446c3e18804021

Intel GPU Base Operator v0.2.1-rc0

Release candidate #0 for 0.2.1.

First public release for Intel GPU Base Operator.

Functionality:

• Deploy either Intel GPU DRA driver or Intel GPU device plugin
• GPU monitoring via Intel XPUManager Daemon (xpumd)
• Scheduling aided with Node Feature Discovery
• Sets up metrics flow to Prometheus
• Allows creating Kueue queues for advanced scheduling
• Initial support for GPU FW update

Basic install with Intel GPU DRA Driver:

helm install --namespace "intel-gpu-operator" --version 0.2.1-rc0   gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart
helm install --namespace "intel-gpu-operator" --version 0.2.1-rc0   gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dra

Basic install with Intel GPU Device Plugin:

helm install --namespace "intel-gpu-operator" --version 0.2.1-rc0   gpu-operator oci://ghcr.io/intel/intel-gpu-base-operator-chart
helm install --namespace "intel-gpu-operator" --version 0.2.1-rc0   gpu-policy oci://ghcr.io/intel/intel-gpu-base-operator-policy-chart --set resourceRegistration=dp

Collateral images:

ghcr.io/intel/intel-resource-drivers-for-kubernetes/intel-gpu-resource-driver:v0.10.0@sha256:746150e64010881dbfdaeb74771703b13cac365a89ee47c4d7499d686ea4163f
ghcr.io/intel/xpumanager/xpumd:v2.0.0-rc.0@sha256:8f020012f68888314402c0332a53718ace4ade9913476bbd125af89edb760a8b
docker.io/intel/intel-gpu-plugin:0.35.0@sha256:34697f9c286857da986381595ac2a693524a83c831955247dae47dfda4d2f526
docker.io/intel/intel-gpu-levelzero:0.35.0@sha256:a8a0729f253de6e8e117a7ef621883a1228f7304747076dcd1446c3e18804021