Compute Engine: NVIDIA P4 GPU shutdown

From 1 Aug 2027, you cannot create, launch, or access any Google Cloud resources that run NVIDIA P4 GPUs. At that date, Google Cloud shuts down any remaining compute instances and resources that run NVIDIA P4 GPUs. Besides Compute Engine instances, the affected resources include Google Kubernetes Engine nodes, Cloud Workstations workstations and Dataflow pipeline jobs. Committed use discounts (CUDs) for NVIDIA P4 GPUs:

• 3-year committed use discounts can no longer be purchased or renewed.
• Existing active 1-year and 3-year commitments remain valid until their expiration date.
• For active commitments that expire after the platform's End of Life (EOL) date, Google says to contact your account team or Technical Account Manager (TAM).

1. Choose a new GPU model. Google recommends the G4 or G2 machine series.
2. Check that the selected GPU model is available in your locations. For services other than Gemini Enterprise Agent Platform and Cloud Workstations, see GPU regions and zones. For those services, see Gemini Enterprise Agent Platform accelerator locations and Cloud Workstations GPU availability by region and zone.
3. For Compute Engine, if the existing instance uses Local SSD disks with data you want to keep, first move the contents to a Persistent Disk volume. Then create a new compute instance. An existing instance cannot be changed in place from an N1 general-purpose machine type to an accelerator-optimized machine type. See Create a G2 or G4 instance.
4. Move Persistent Disk volumes to the new instance, reassign static IP addresses, and install GPU drivers and your applications on it.
5. Delete the old compute instance.
6. For other Google Cloud services, update your configurations to reference a supported GPU model, then restart or recreate your resources.