[GCP] Cloud TPU
Cloud GCP
Published: 2020-12-01

Cloud TPU

Cloud TPU is Google’s machine learning processor, and many companies may be considering installing it to analyze their customer data. In this article, I will introduce how to use Cloud TPU and its advantages and disadvantages. If you want to perform machine learning, please refer to this article.

How to use Cloud TPU

To use Cloud TPU, you can configure various settings from Cloud Shell.

First, let’s open the Cloud Shell window.

What you will do in Cloud Shell is as follows.

  • Create a variable for the project ID
  • Deploy the gcloud command line tool
  • Create a Cloud Storage bucket
  • Start the Compute Engine VM and Cloud TPU

Each of them has a specific command, so copy and paste it into Cloud Shell and run it.

Next, download the source code for the MNIST TPU model. After setting the environment variables and running the MNIST training script, the minimum preparation is complete.

Advantages of Cloud TPU

Lower initial investment.

With the ability to use TPU in the cloud, you can perform the latest machine learning with low initial investment. There is no need to make a huge investment to create an AI as in the past.

Fast computation

Cloud TPU is a processor with improved computing performance instead of versatility. It can drastically reduce learning time, so you can get more accurate analysis results with more trials. What would have taken a week or so with other machine learning processors can be completed in about an hour with Cloud TPU.

Easy to use, even for beginners

Cloud TPU has developed test models for different genres such as image and speech recognition. These models are open source and free, so even beginners can easily perform highly accurate machine learning.

Disadvantages of Cloud TPU

Due to the configuration of the system, Cloud TPU is not capable of computing all data. Some of the data cannot be analyzed, so you need to check when using it. For example, data sets that require access to memory are not suitable for Cloud TPU. On the other hand, if you are mainly using matrix calculations, you may be able to realize the benefits of Cloud TPU.


Cloud TPU allows you to perform modern machine learning at a low cost. Because of its high computing power, it is also recommended for analyzing huge amounts of data. Open source test models have been developed, so even beginners should actively use them.

Related Articles