Cloud TPU machine learning accelerators: Now available to you in beta

Feb. 13, 2018, 9:12 p.m. By: Kirti Bakshi

Cloud TPU

Starting as of now, Cloud TPUs are now made to be available in beta on Google Cloud Platform (GCP) in order to help machine learning (ML) experts train and run their ML models more quickly and efficiently. But before we go ahead into the same in depth:

What actually are Cloud TPU's?

Talking of Cloud TPU's, Cloud TPUs are a family of hardware accelerators that have been Google-designed and are optimized to speed up as well as scale up specific ML workloads that are programmed with TensorFlow. Since it has been built with custom ASICs that are four in number, each Cloud TPU packs up to 64 GB of high-bandwidth memory onto a single board and 180 teraflops of floating-point performance. These boards can be used alone or connected together with the help of an ultra-fast, dedicated network to form multi-petaflop ML supercomputers that are called as “TPU pods” These larger supercomputers on GCP will be offered later this year.

Main Motive Behind The Design:

The main motive behind designing Cloud TPUs was to deliver for targeted TensorFlow workloads, differentiated performance per dollar and also alongside to enable ML engineers and researchers to iterate the same more quickly.

Machine Learning model training, now made easier:

Writing programs traditionally, for custom ASICs and supercomputers before used to require deeply specialized expertise. By contrast, one can now program Cloud TPUs with high-level TensorFlow APIs, and the team has open-sourced a set of reference high-performance Cloud TPU model implementations to help you get started right away:

  • ResNet-50 and other popular models for image classification.

  • Transformer for machine translation and language modelling.

  • RetinaNet for object detection.

To save you time and effort, the team continuously tests these model implementations both for performance and for convergence to the expected accuracy on standard datasets.

Over time, it will also open-source additional model implementations. As a result of which, ML experts who are adventurous in nature may be able to optimize other TensorFlow models for Cloud TPUs on their own by making the use of the documentation and tools that are provided.

An ML platform that is scalable:

Cloud TPU's also simplify planning and managing ML computing resources:

  • You can provide your teams with state-of-the-art ML acceleration and adjust your capacity dynamically as their needs change.

  • As an alternative of committing to the capital, time as well as expertise that is required to design, install and maintain an on-site ML computing cluster with specialized power, cooling, networking and storage requirements, you can benefit from what has been heavily optimized at Google over many years: A large-scale, tightly-integrated ML infrastructure.

  • There’s no more struggling to keep drivers up-to-date across a large collection of workstations and servers. Cloud TPUs are preconfigured—no driver installation required!

  • You are protected by the same sophisticated security mechanisms and practices that safeguard all Google Cloud services.

Getting started with Cloud TPUs:

Cloud TPUs are as of now available in limited quantities and usage is billed by the second at the rate of $6.50 USD / Cloud TPU / hour.

As of now, there has been great enthusiasm from customers as they expressed for Cloud TPUs. In order to help them manage demand, one can sign up to the below-mentioned link to request Cloud TPU quota and describe your various ML needs. The best will be done to give you access to Cloud TPUs as soon as it is possible.

In order to learn more about Cloud TPUs, join the same for a Cloud TPU webinar that will be held on the 27th of February, 2018. The links to the same have been provided below:

For More Information: Click Here.