How to create a TPU VM in Google Cloud console (GUI)

[hugohuangtao]

Hey everyone,

While CLI commands are great for automation, launching a TPU VM via the Google Cloud Console UI is often the easiest way to visually inspect available machine types, pick the right accelerator generation, and select operating system images.

Below is a complete walkthrough on how to provision a modern TPU VM (such as v5e, v6e, or TPU7x) using the Google Cloud Console.

Here is a visual, step-by-step Discourse post designed to guide users through the Google Cloud console interface, complete with screenshot placeholders and practical tips.

---

Title: [Step-by-Step Guide] How to Create a TPU VM in the Google Cloud Console
Category: Cloud Infrastructure / Tutorials
Tags: cloud-tpu, compute-engine, gcp-console, tutorial, machine-learning

---

Hey everyone,

While CLI commands are great for automation, launching a TPU VM via the Google Cloud Console UI is often the easiest way to visually inspect available machine types, pick the right accelerator generation, and select operating system images.

Below is a complete walkthrough on how to provision a modern TPU VM (such as v5e, v6e, or TPU7x) using the Google Cloud Console.

---

Prerequisites Before Launching

1. TPU Quota: Ensure your project has allocated quota in your target region (e.g., us-east5-a, us-west4-a, or europe-west4-a).

2. Compute Engine API: Enabled on your target project.

---

Step 1: Navigate to VM Instances

1. Open the Google Cloud Console.

2. In the navigation menu (top-left burger menu ☰), go to Compute Engine > VM instances.

3. Click the Create instance button at the top toolbar.

Step 2: Name and Placement (Region & Zone)

1. Name: Provide a lowercase, DNS-compliant name for your instance (e.g., tpu-v6e-worker-01).

2. Region & Zone: Select a region and zone that hosts TPU hardware for your desired generation (e.g., us-east5-a for v6e, us-west4-a for v5e).

Step 3: Select Machine Family & TPU Generation

Under the Machine configuration section:

1. Select the TPUs (or Accelerator-optimized) tab in the machine family picker.

2. In the Series table, select your desired architecture:

   • v6e (Trillium / AMD EPYC)

   • v5e / v5p

   • TPU7x (Ironwood)

3. Choose your Machine type depending on the chip density you need:

   • e.g., ct6e-standard-1t (1 chip), ct6e-standard-4t (4 chips), or ct5lp-hightpu-1t.

Step 4: Configure the OS Image and Storage

Click on the left of the screen: the OS and storage section:

1. Under the Operating system and storage:

   • Image: Select Ubuntu or specialized accelerator image families.

   • Version: In our case, the system selected Ubuntu 22.04 LTS (Jammy) for you.

2. Size: Choose Balanced Persistent Disk or SSD Persistent Disk (recommended: at least 100 GB to hold ML datasets, Docker images, and dependencies).

3. Click Select.

Step 5: Review & Click Create

1. Review the estimated monthly and hourly costs shown in the right-hand summary pane.

2. (Optional Pro-Tip) Click Equivalent code at the bottom if you want to inspect or copy the exact gcloud or terraform syntax generated by the UI.

3. Click Create.