This Month in GKE: June Edition

GKE Updates - June 2025 Edition

Key Concepts: GKE (Google Kubernetes Engine), Autopilot, Standard Mode, Compute Class, Flex Starts, TPU (Tensor Processing Unit), MCP (Model Context Protocol), kubectl-ai, Site Care Containers, Backup for GKE, Cross-Project Backup/Restore, C4D VMs, AMD EPYC Turin Processor, Google Titanium, Custom Compute Class, Zonal Preferences, Placement Policy, Compact Placement Policy, HPC (High-Performance Computing), Cost Optimization Dashboard, Cost Allocation, Extended Memory.

Autopilot Compute Class in GKE Standard (Private Preview)

• Main Point: Users can now leverage Autopilot features within GKE Standard mode without choosing between operating modes.
• Details: Enables running workloads in an Autopilot, secure user space environment.
• Benefits: Reduced operational overhead and automated capacity provisioning.

Flex Starts with TPU Support on GKE

• Main Point: Flex Starts now supports TPUs in single-host node pools.
• Details: Flex Starts is a dynamic workload scheduler feature.
• Benefits: Easier access to high-demand accelerators like TPU v5e, v5p, and Trillum while optimizing utilization.

GKE MCP Server Release

• Main Point: Release of a GKE MCP (Model Context Protocol) server.
• Functionality: Allows plugging in tools to AI assistance platforms like Claude, Cursor, and Gemini CLI.
• Current Capabilities:
  • Listing all clusters in a project.
  • Retrieving specific cluster information.
  • Generating manifests for inference workloads (e.g., running Llama on vLLM).
• Integration: Works with the open-source tool kubectl-ai.
• Example: Generates a manifest for running Llama on vLLM.

CPU and Memory Request/Limit Metrics for Site Care Containers

• Main Point: GKE now reports CPU and memory requests and limits metrics for Kubernetes native site care containers.
• Version Requirement: Requires GKE version 1.32.0.4 and above.

Backup for GKE: Cross-Project Backup and Restore

• Main Point: Backup for GKE now supports cross-project backups and restores.
• Status: Generally Available (GA) but requires allow listing.
• Action: Contact your Google account team to enable it for your project.

C4D Virtual Machines (Generally Available)

• Main Point: C4D virtual machines are now generally available in GKE.
• Version Requirements: GKE Standard (version 1.32), GKE Autopilot (version 1.33).
• Hardware: Powered by the 5th generation AMD EPYC Turin processor and Google's Titanium silicon.

Custom Compute Class Enhancements

• Zonal Preferences: Users can specify a zone with reconciliation to a preferred zone for each priority.
• Placement Policy: Supports compact placement policy.
• Use Case: Important for HPC (High-Performance Computing) applications.

GKE Cost Optimization Dashboard Recommendations

• Main Point: The GKE Cost Optimization Dashboard provides new recommendations.
• Insights: Identifies idle, over-provisioned, and under-provisioned resources.
• Metrics: Displays the dollar value associated with these resource inefficiencies.
• Accessibility: Insights are visible directly within the Google Cloud Console.

GKE Cost Allocation: Extended Memory Support

• Main Point: GKE Cost Allocation now supports extended memory.
• Problem Addressed: Previously, the cost of extended memory on certain VM types was not properly allocated to specific workloads or namespaces.
• Solution: The full cost of extended memory is now allocated to the correct workload or namespace.

Conclusion

The June 2025 GKE update brings several key enhancements focused on flexibility, AI integration, backup capabilities, performance, and cost optimization. The introduction of Autopilot compute class in GKE Standard, Flex Starts support for TPUs, and the GKE MCP server significantly expand the platform's capabilities. Improvements to backup and restore, the availability of C4D VMs, and enhancements to cost allocation provide users with more control, performance, and cost efficiency.