ack-nvidia-device-plugin release notes - Container Service for Kubernetes

The NVIDIA Device Plugin is a Kubernetes component that manages GPUs on each node, helping Kubernetes use GPU resources more efficiently. This topic covers usage and release notes.

Usage

Configure and manage NVIDIA Device Plugin covers version management, upgrades, restarts, and device isolation.

Version differences

The ack-nvidia-device-plugin add-on is deployed and managed differently depending on your cluster version. If your cluster version is earlier than 1.20, manually upgrade the cluster before proceeding.

Attribute	Cluster versions 1.32 and later	Cluster versions 1.20 to 1.31
Deployment method	DaemonSet	Static pod
Management method	Add-on management in the console	Manual maintenance
Plugin upgrade on node pool upgrade	No (upgrade the add-on separately)	Yes (automatic)
Node label requirements	`ack.node.gpu.schedule=default`	No special requirements

Release notes

Clusters v1.32 and later

May 2026

Version	Image tag	Update date	Description	Impact
v0.7.0	v0.19.1-ack-1.0-2fd61386	May 15, 2026	Based on community NVIDIA Device Plugin v0.19.1. GPU auto-isolation trigger changes from enabled by default to enabled via configuration. See GPU health check and auto drain. See [Product Update] Announcement on Changes to the GPU Auto-Isolation Feature. Health check now validates extended resource counts in kubelet. Sends a `NvidiaGPUResourceCountMismatch` event when the `nvidia.com/gpu` resource count is incorrect.	No workload impact.

November 2025

Version	Image tag	Update date	Description	Impact
v0.5.0	v0.18.0-ack-1.0-2ac7c6b3	November 25, 2025	Based on community NVIDIA Device Plugin v0.18.0. Adds GDRCopy support. Fixes probe timeout and restart on nodes with many pods.	No workload impact.

September 2025

Version	Image tag	Update date	Description	Impact
v0.4.0	v0.17.3-ack-1.0-b70733ec	September 16, 2025	Updates health check and changes check interval to 8 minutes. Based on community NVIDIA Device Plugin v0.17.3.	No workload impact.

July 2025

Version	Image tag	Update date	Description	Impact
v0.3.0	v0.17.0-ack-1.0-678014dd	July 14, 2025	Adds node label support for enabling CDI. Based on community NVIDIA Device Plugin v0.17.0.	No workload impact.

April 2025

Version	Image tag	Update date	Description	Impact
v0.2.0	v0.17.0-49a827fd-aliyun	April 29, 2025	Based on community NVIDIA Device Plugin v0.17.0. Adds 5-minute health check to prevent nodes from reporting zero GPU resources when the gRPC connection between device plugin and kubelet disconnects. Fixes continuous GPU pod creation retries on GPU card failure. Disables device isolation by default on node XID errors.	No workload impact.

December 2024

Version	Image tag	Update date	Description	Impact
v0.1.0	v0.17.0-1-aliyun	December 26, 2024	Based on community NVIDIA Device Plugin v0.17.0.	No workload impact.

Clusters v1.20 to v1.31

January 2025

Image tag	Update date	Description	Impact
v0.9.3-0dd4d5f5-aliyun	January 16, 2025	Enables device isolation by default on node XID errors. Changes the CPU and memory values in `resources.limits` and `resources.requests`.	No workload impact.

April 2024

Image tag	Update date	Description	Impact
v0.9.1-576cfc77-aliyun	April 8, 2024	Removes the `livenessProbe`.	No workload impact.

February 2024

Image tag	Update date	Description	Impact
v0.9.1-3f942982-aliyun	February 29, 2024	Adds a `livenessProbe`.	No workload impact.

August 2023

Image tag	Update date	Description	Impact
v0.9.0-70671340-aliyun	August 2, 2023	Modifies `resources.limits`.	No workload impact.

April 2021

Image tag	Update date	Description	Impact
1.12	April 1, 2021	Customizes GPU device mounting logic.	No workload impact.