Cloud TPU pricing

TPU pricing and quota are divided into two systems:

• Single device TPU type pricing for individual TPU devices that are available either on-demand or as preemptible devices. You cannot combine multiple single device TPU types to collaborate on a single workload.
• TPU Pod type pricing for clusters of TPU devices that are connected to each other over dedicated high-speed networks. These TPU types are available only if you have evaluation quota or purchase a 1 year or 3 year commitment.

Charges for Cloud TPU accrue while your TPU node is in a READY state. You receive a bill at the end of each billing cycle that lists usage and charges for that billing cycle.

To learn how to request quota for single device TPU types, read the Cloud TPU quota policy page.

Important: Charges for your TPU nodes stop only when you stop or delete the TPU node. Stopping or deleting a virtual machine that you created with the ctpu utility does NOT stop billing on the TPU node. See Creating and deleting TPUs to learn how to stop and delete a TPU node.

Note: Cloud TPU offers promotional pricing to all customers specifically for the v2-8 TPU type. The v2-8 information on this page shows the reduced pricing.

Single device pricing

Single device TPU types are billed in one-second increments and available at either an on-demand or preemptible price.

Single device TPU types are independent TPU devices direct network connections to other TPU devices in a Google data center. If your workloads require more TPU cores and a larger pool of memory, use a TPU Pod type instead.

A preemptible TPU is one that Cloud TPU can terminate (preempt) at any time if Cloud TPU requires access to the resources for another task. The charges for a preemptible TPU are much lower than those for a normal TPU. You are not charged for preemptible TPUs if they are preempted in the first minute after you create them.

You can configure your TPU nodes with the following single device TPU types:

Iowa (us-central1) Netherlands (europe-west4) Taiwan (asia-east1)

Monthly

Hourly

TPU type (v2)	v2 cores	Total memory	On-demand price (USD)	Preemptible price (USD)
v2-8	8	64 GiB
TPU type (v3)	v3 cores	Total memory	On-demand price (USD)	Preemptible price (USD)
v3-8	8	128 GiB

If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.

TPU Pod type pricing (Beta)

Beta: Cloud TPU Pods

This is a Beta release of Cloud TPU v2 and v3 Pods. This release is not covered by any SLA or deprecation policy and is not intended for real-time use in critical applications.

TPU Pod types provide access to multiple TPU devices that are all connected on a dedicated high-speed network. These TPU types provide greater compute capacity and a larger pool of TPU memory to a single TPU node. To use TPU Pod types, you must request quota using one of the following options:

• Request access to evaluation quota so that you can test the performance of TPU Pod types. TPU nodes that you create using evaluation quota are billed in one-second increments but do not guarantee the same level of service as on-demand TPU devices or devices that you create using commitment quota. Evaluation quota persists only for a limited amount of time on your project.
• Purchase a 1 year or 3 year commitment and create TPU nodes with up to 2048 cores. Commitments are not billed incrementally. Commitments bill you a monthly fee for the duration of your commitment term even if you do not use any TPU resources.

You can configure your TPU nodes with the following TPU types:

Iowa (us-central1) Netherlands (europe-west4) Taiwan (asia-east1)

TPU type (v2)	v2 cores	Total memory	Evaluation price (USD)	1-year commitment price (USD)	3-year commitment price (USD)
v2-32 (Beta)	32	256 GiB
v2-128 (Beta)	128	1 TiB
v2-256 (Beta)	256	2 TiB
v2-512 (Beta)	512	4 TiB
TPU type (v3)	v3 cores	Total memory	Evaluation price (USD)	1-year commitment price (USD)	3-year commitment price (USD)
v3-32 (Beta)	32	512 GiB
v3-64 to v3-2048 (Beta)	64 to 2048	1 TiB to 32 TiB	Contact sales to request pricing for these larger v3 TPU types.

If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.

To learn about the differences between different TPU versions and configurations, read the TPU System Architecture documentation.

Total cost for 1 year and 3 year commitments

The following table shows the total cost for various TPU commitments during their full duration:

Iowa (us-central1) Netherlands (europe-west4) Taiwan (asia-east1)

TPU type (v2)	1-year commitment cost	3-year commitment cost
v2-32 (Beta)
v2-128 (Beta)
v2-256 (Beta)
v2-512 (Beta)
TPU type (v3)	1-year commitment cost	3-year commitment cost
v3-32 (Beta)
v3-64 to v3-2048 (Beta)	Contact sales to request pricing for these larger v3 TPU types.

If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.

To request access to TPU types with more than 8 cores, contact a sales representative.

TensorFlow Research Cloud (TFRC) program

If you're enrolled in the TFRC program you are granted access to Cloud TPU v2 and v3 for a limited period of time free of charge. Specifically for the TFRC program, you are not charged for Cloud TPU as long as your TPU nodes run in the us-central1-f zone.

Virtual machine pricing

In order to connect to a TPU, you must provision a virtual machine (VM), which is billed separately. For details on pricing for VM instances, see Compute Engine pricing.

Note: Charges for VM and Cloud TPU usage are separate. Charges start for Cloud TPU only when the TPU is started and charges stop when the TPU is stopped or deleted. Stop the TPU by running ctpu pause or gcloud compute tpus stop. Similarly, you are only charged for the VM when the VM is active. If the VM is stopped and the Cloud TPU is not, you continue to be charged for the Cloud TPU. If the Cloud TPU is stopped or deleted and the VM is not, then you continue to be charged for the VM.

Pricing calculator

To estimate the cost of using Cloud TPU with Compute Engine VM instances, see the Compute Engine pricing calculator.

Pricing example

The following example explains how to calculate the total cost of a training job that uses TPU resources and Compute Engine instances in the US region.

A machine learning researcher provisions a virtual machine by creating a Compute Engine instance, and they select the n1-standard-2 machine type. They also create a TPU resource, and they accrue 10 hours of usage on both the Compute Engine instance and the TPU resource. In order to calculate the total cost of a training job, the machine learning researcher must add together:

• the total cost of all Compute Engine instances
• the total cost of all Cloud TPU resources

Resource	Price per machine per hour in USD	Number of machines	Number of hours billed	Total cost of each resource	Total cost of training job
Compute Engine n1-standard-2 instance	$0.095	1	10	$0.95
Cloud TPU resource	$4.50	1	10	$45.00
					$45.95

Pricing example using a preemptible TPU

The following example uses the same resources and time period as above, except that the researcher decides to use a preemptible TPU to save costs. The US charge for the preemptible TPU v2 is $1.35 USD per hour, as opposed to $4.50 for a normal TPU v2 .

Resource	Price per machine per hour in USD	Number of machines	Number of hours billed	Total cost of each resource	Total cost of training job
Compute Engine n1-standard-2 instance	$0.095	1	10	$0.95
Preemptible TPU	$1.35	1	10	$13.50
					$14.45

What's next

• Read the Cloud TPU quota policy page to learn how to request quota for different TPU types.
• Estimate the cost of using Cloud TPU with the Compute Engine pricing calculator.
• Check the regions and zones in which Cloud TPU is available.
• Check the release notes for future updates on pricing.