Release Notes

This page documents production updates to Cloud TPU. You can periodically check this page for announcements about new or updated features, bug fixes, known issues, and deprecated functionality.

You can see the latest product updates for all of Google Cloud on the Google Cloud page, browse and filter all release notes in the Google Cloud Console, or you can programmatically access release notes in BigQuery.

To get the latest product updates delivered to you, add the URL of this page to your feed reader, or add the feed URL directly: https://cloud.google.com/feeds/tpu-release-notes.xml

December 02, 2021

Cloud TPU team just released TF-2.4.4, TF-2.5.2 and TF-2.6.2 on Cloud TPUs. The TensorFlow release notes for these releases are shown below.

November 05, 2021

Cloud TPU now supports Tensorflow 2.7.0. For more information, see Tensorflow 2.7.0 Release Notes

August 24, 2021

Cloud TPU team just released TF-2.3.4, TF-2.4.3 and TF-2.5.1 on Cloud TPUs. The TensorFlow release notes for these releases are shown below.

August 12, 2021

Cloud TPU now supports Tensorflow 2.6.0. For more information, see Tensorflow 2.6.0 Release Notes.

In TF 2.6.0, TensorFlow has introduced a new version of the TF/XLA bridge using the MLIR compiler infrastructure. The MLIR bridge is enabled by default. To explicitly disable it at runtime, add the following code snippet to your model's code:

tf.config.experimental.disable_mlir_bridge()

June 22, 2021

The Cloud TPU team has released support for TensorFlow 2.4.2. The corresponding Tensorflow release notes are:

Tensoflow-2.4.2 Release notes

June 17, 2021

Cloud TPU team just released TF-2.1.4, TF-2.2.3 and TF-2.3.3 on Cloud TPUs. The TensorFlow release notes for these releases are shown below.

June 07, 2021

Cloud TPU now supports Tensorflow 2.5.0. For more information, see Tensorflow 2.5.0 Release Notes

June 01, 2021

New Cloud TPU VMs make training your ML models on TPUs easier than ever

The new Cloud TPU VM architecture makes it easier than ever before to use our industry-leading TPU hardware. The Cloud TPU VMs provide direct access to TPU host machines, offering a new and improved user experience for developing and deploying TensorFlow, PyTorch, and JAX on Cloud TPUs. Instead of accessing Cloud TPUs remotely over the network, Cloud TPU VMs let you set up your own interactive development environment on each TPU host machine. Now you can write and debug an ML model line-by-line using a single TPU VM, and then scale it up on a Cloud TPU Pod slice to take advantage of the super-fast TPU interconnects. You have root access to every TPU VM you create, so you can install and run any code you wish in a tight loop with your TPU accelerators. You can use local storage, execute custom code in your input pipelines, and more easily integrate Cloud TPUs into your research and production workflows. Google supports Cloud TPU integrations with TensorFlow, PyTorch, and JAX, and you can even write your own integrations via a new libtpu shared library on the VM. For more information, see https://cloud.google.com/blog/products/compute/introducing-cloud-tpu-vms.

December 14, 2020

Cloud TPU now supports Shared VPC

Shared VPC allows an organization to connect resources from multiple projects to a common VPC network to communicate with each other securely and efficiently using internal IPs from that network. This release enables connecting to Cloud TPU Nodes from Shared VPC networks.

September 11, 2020

Compute Engine TPU Metrics and Logs In-Context

New Monitoring tab for TPUs provides key TPU Metrics and access to logs at a glance. You can see a variety of key TPU metrics including MXU utilization, CPU, memory, sent/received traffic, and more. In addition, it provides instant access to TPU logs which give insight into important events.

August 20, 2020

PyTorch/XLA 1.6 Release (GA)

Highlights

Cloud TPUs now support the PyTorch 1.6 release, via PyTorch/XLA integration. With this release we mark our general availability (GA) with the models such as ResNet, FairSeq Transformer and RoBERTa, and HuggingFace GLUE task models that have been rigorously tested and optimized.

In addition, with our PyTorch/XLA 1.6 release, you no longer need to run the env-setup.py script on Colab/Kaggle as those are now compatible with native torch wheels. You can still continue to use that script if you would like to run with our latest unstable releases.

New Features

  • XLA RNG state checkpointing/loading (https://github.com/pytorch/xla/pull/2096)
  • Device Memory XRT API (https://github.com/pytorch/xla/pull/2295)
  • [Kaggle/Colab] Small host VM memory environment utility (https://github.com/pytorch/xla/pull/2025)
  • [Advanced User] XLA Builder Support (https://github.com/pytorch/xla/pull/2125)
  • New op supported on PyTorch/XLA
    • Hardsigmoid (https://github.com/pytorch/xla/pull/1940)
    • true_divide (https://github.com/pytorch/xla/pull/1782)
    • max_unpool2d (https://github.com/pytorch/xla/pull/2188)
    • max_unpool3d (https://github.com/pytorch/xla/pull/2188)
    • Replication_pad1d (https://github.com/pytorch/xla/pull/2188)
    • Replication_pad2d (https://github.com/pytorch/xla/pull/2188)
  • Dynamic shape support on XLA:CPU and XLA:GPU (experimental)

Bug Fixes

  • RNG Fix (proper dropout)
  • Manual all-reduce in backward pass (https://github.com/pytorch/xla/pull/2325)

August 19, 2020

Cloud TPU now supports Shared VPC in Beta.

Shared VPC allows an organization to connect resources from multiple projects to a common VPC network to communicate with each other securely and efficiently using internal IPs from that network. This release enables connecting to Cloud TPU Nodes from Shared VPC networks.

May 29, 2020

Cloud TPU now supports TensorFlow version 1.15.3. See the TensorFlow 1.15.3 Release Notes.

May 21, 2020

Cloud TPU now supports TensorFlow 2.1.1 with Keras support. See the TensorFlow 2.1.1 Release Notes for a complete list of features included in this release.

May 12, 2020

Cloud TPU currently supports TensorFlow version 1.15.2. See the Release Notes.

TensorFlow 1.15 supported Python 2, but that support has been discontinued with TensorFlow 1.15.2.

May 08, 2020

Cloud TPU now supports TensorFlow 2.2. See the TensorFlow 2.2 Release Notes for a complete list of features included with this release. New models for Image segmentation and Image classification have been added to the official cloud TPU supported models list.

April 21, 2020

Cloud TPUs and Cloud TPU Pods now support PyTorch 1.5 via the PyTorch/XLA integration. This integration makes it possible for PyTorch users to do everything they can do on GPUs on Cloud TPUs, while minimizing changes to the user experience. You can try out PyTorch on an 8-core Cloud TPU device for free via Google Colab, and you can use PyTorch on Cloud TPUs at a much larger scale on Google Cloud (all the way up to full Cloud TPU Pods).

See the ⁠PyTorch/XLA 1.5 Release Notes for a complete list of features included in this release.

January 09, 2020

Cloud TPU now supports TensorFlow 2.1 with Keras support. See the TensorFlow 2.1 Release Notes for a complete list of features included in this release.

December 05, 2019

Cloud TPU v2 and v3 Pods are now Generally Available for TensorFlow version 1.x. Supported models can be found here.

Since TPU resources can scale from a single Cloud TPU to a Cloud TPU Pod, you don't need to choose between a single Cloud TPU and a Cloud TPU Pod. You can request portions of Cloud TPU Pods in slices or sets of cores, so that you purchase only the processing power you need.

Cloud TPU v2 and v3 Pod advantages over a single v2 or v3 Cloud TPU device:

  • Increased training speeds for fast iteration in R&D *Increased human productivity by providing automatically scalable machine learning (ML) compute
  • Ability to train much larger models

Cloud TPU v3 Pod advantages over Cloud TPU v2 Pod:

  • Faster processing and larger memory:
    • v2 Pod: 11.5 petaflops and 4 TB on-chip memory (HBM)
    • v3 Pod: 100 petaflops and 32 TB HBM, with liquid cooling* Can train even larger models

October 22, 2019

Cloud TPU now supports TensorFlow version 1.15 (Release Notes, API Documentation). See the current supported TensorFlow versions in the Cloud TPU supported versions document.

Cloud TPU support for TensorFlow 1.15 includes the following changes:

July 29, 2019

Cloud TPU now supports TensorFlow version 1.14. Support for Tensorflow versions 1.11 is removed. See the current supported TensorFlow versions in the Cloud TPU versioning policy.

Cloud TPU support for TensorFlow 1.14 includes the following changes:

  • Improved Error Messages: Cloud TPU errors in TensorFlow 1.14 are aggregated across multiple TPU cores and across multiple workers. This change makes error messages more comprehensible for user code.
  • Redesigned object detection codebase: The object detection codebase provides optimized training performance, clean and configurable parameter management, and advanced features such as spatial partition, NAS-FPN and AutoAugment.

May 07, 2019

Cloud TPU v2 Pod is available in Beta release.

Since TPU resources can scale from a single Cloud TPU to a Cloud TPU Pod, you don't need to choose between a single Cloud TPU and a Cloud TPU Pod. You can request portions of Cloud TPU Pods in slices or sets of cores, so that you purchase only the processing power you need.

Cloud TPU Pod (beta) advantages over a single Cloud TPU v2 device:

  • Increased training speeds for fast iteration in R&D
  • Increased human productivity by providing automatically scalable machine learning (ML) compute
  • Ability to train much larger models

Cloud TPU v3 Pod is available in Beta release.

Since TPU resources can scale from a single Cloud TPU to a Cloud TPU Pod, you don't need to choose between a single Cloud TPU and a Cloud TPU Pod. You can request portions of Cloud TPU Pods in slices or sets of cores, so that you purchase only the processing power you need.

Cloud TPU Pod (beta) advantages over a single v3 Cloud TPU device:

  • Increased training speeds for fast iteration in R&D
  • Increased human productivity by providing automatically scalable machine learning (ML) compute
  • Ability to train much larger models

Cloud TPU v3 Pod (beta) advantages over Cloud TPU v2 Pod (beta):

  • Faster processing and larger memory:
    • v2 Pod: 11.5 petaflops and 4 TB on-chip memory (HBM)
    • v3 Pod: 100 petaflops and 32 TB HBM, with liquid cooling* Can train even larger models

March 11, 2019

Cloud TPU now supports TensorFlow version 1.13. Support for Tensorflow versions 1.8 and 1.9 have been removed.

See the current supported TensorFlow versions in the Cloud TPU versioning policy.

January 31, 2019

Cloud TPU v3 is now GA (generally available). Cloud TPU v3 has double the memory of v2. This gives improved performance and enables support for more classes of models, for example deeper ResNets and larger images with RetinaNet. Existing models that run on Cloud TPU v2 will continue to work. Refer to the Cloud TPU versions guide for more information.

November 08, 2018

Cloud TPU now supports TensorFlow version 1.12. This release includes improvements for Keras on Cloud TPUs, performance optimizations throughout the software stack, and improved APIs, error messages, and reliability.

See the current supported TensorFlow versions in the Cloud TPU versioning policy.

November 07, 2018

Cloud TPU v2 Pod is available in Alpha release.

Since TPU resources can scale from a single Cloud TPU to a Cloud TPU Pod, you don't need to choose between a single Cloud TPU and a Cloud TPU Pod. You can request portions of Cloud TPU Pods in slices or sets of cores, so that you purchase only the processing power you need.

Cloud TPU Pod (alpha) advantages:

  • Increased training speeds for fast iteration in R&D
  • Increased human productivity by providing automatically scalable machine learning (ML) compute
  • Ability to train much larger models than on a single ML accelerator

October 10, 2018

Cloud TPU v3 is available in Beta release. You now have a choice between v2 and v3 in your configuration.

  • Cloud TPU v3 has double the memory of v2. This gives improved performance and enables support for more classes of models, for example deeper ResNets and larger images with RetinaNet.
  • Existing models that run on Cloud TPU v2 will continue to work.
  • Refer to the Cloud TPU versions guide for more information.

Preemptible TPUs are now GA (generally available). A preemptible TPU is a Cloud TPU node that you can create and run at a much lower price than normal nodes. However, Cloud TPU may terminate (preempt) these nodes if it requires access to the resources for another purpose.

September 27, 2018

Cloud TPU now supports TensorFlow version 1.11. TensorFlow 1.11 introduces experimental support for all of the following on Cloud TPU: Keras, Colab, eager execution, LARS, RNNs, and Mesh TensorFlow. This release also introduces a high-performance Cloud Bigtable integration, new XLA compiler optimizations, other performance optimizations throughout the software stack, and it provides improved APIs, error messages, and reliability.

See the current supported TensorFlow versions in the Cloud TPU versioning policy.

September 07, 2018

Support for TensorFlow version 1.7 ended on September 7, 2018. See the current supported versions in the Cloud TPU versioning policy.

July 24, 2018

We're delighted to announce promotional pricing for Cloud TPU, resulting in significant price reductions. The following table shows the previous pricing and the new pricing available from today:

US

Previous price per TPU per hour New price per TPU per hour
Cloud TPU $6.50 USD $4.50 USD
Preemptible TPU $1.95 USD $1.35 USD

Europe

Previous price per TPU per hour New price per TPU per hour
Cloud TPU $7.15 USD $4.95 USD
Preemptible TPU $2.15 USD $1.485 USD

Asia Pacific

Previous price per TPU per hour New price per TPU per hour
Cloud TPU $7.54 USD $5.22 USD
Preemptible TPU $2.26 USD $1.566 USD

See the pricing guide for details.

July 12, 2018

Cloud TPU is now available in Google Kubernetes Engine as a Beta feature. Run your machine learning workload in a Kubernetes cluster on GCP, and let GKE manage and scale the Cloud TPU resources for you.

  • Follow the tutorial to train the Tensorflow ResNet-50 model on Cloud TPU and GKE.
  • Refer to the GKE setup guide for quick instructions on running Cloud TPU with GKE.

July 02, 2018

Cloud TPU now supports TensorFlow version 1.9. TensorFlow 1.9 brings increases in Cloud TPU performance as well as improved APIs, error messages, and reliability.

June 27, 2018

Cloud TPU is now GA (generally available). Google's revolutionary TPUs are designed to accelerate machine learning workloads with TensorFlow. Each Cloud TPU provides up to 180 teraflops of performance, providing the computational power to train and run cutting-edge machine learning models.

June 18, 2018

Preemptible TPUs are now available in Beta. A preemptible TPU is a Cloud TPU node that you can create and run at a much lower price than normal nodes. However, Cloud TPU may terminate (preempt) these nodes if it requires access to the resources for another purpose.

Cloud TPU is now available in the European (EU) and Asia Pacific (APAC) regions as well as the United States (US). See the the pricing details per region. The following zones are available:

  • US
    • us-central1-b
    • us-central1-c
    • us-central1-f (TFRC program only)
  • EU
    • europe-west4-a
  • APAC
    • asia-east1-c

June 12, 2018

Support for TensorFlow version 1.6 ended on June 12, 2018. See the current supported versions in the Cloud TPU versioning policy.

April 20, 2018

Cloud TPU now supports TensorFlow version 1.8. TensorFlow 1.8 brings increases in Cloud TPU performance as well as improved APIs, error messages, and reliability.

Support for TensorFlow version 1.7 ends on June 20, 2018. See the details in the Cloud TPU versioning policy.

April 02, 2018

Cloud TPU now supports TensorFlow version 1.7. Support for TensorFlow version 1.6 ends on June 2, 2018. See the details in the Cloud TPU versioning policy.

February 12, 2018

Cloud TPU is available in Beta release. Google's revolutionary TPUs are designed to accelerate machine learning workloads with TensorFlow. Each Cloud TPU provides up to 180 teraflops of performance, providing the computational power to train and run cutting-edge machine learning models.