Google Cloud Apps Admins

Accelerate Machine Learning Workloads with Google's New Chip off the Old Block.

Written by Christopher Alghini (Google Cloud Consultant) | February 23, 2018

Over the years, we've seen some incredible growth in the world of machine learning. The neural networks that have been developed by technology experts have dramatically enhanced the quality of the Google Translate system, playing a significant role in the way we search for things online. The more insightful the machine learning world becomes, the more we learn about the true possibilities of artificial intelligence.

Now, in an effort to support the path to AI, Google Cloud has announced that they will be making TPUs available in their Google Cloud Platform. TPU stands for Tensor Processing Unit, and it's a processing system designed for the Google open-source framework of machine learning. The specialist chips in TPUs can provide up to 180 teraflops of processing power to support the complex needs of machine learning algorithms. In fact, these are the systems that have been powering Google Data centers for the last few years.

Currently, Google Cloud TPUs are only available in the beta channel. However, Google does plan on adding additional open-source model implementations over time. Machine Learning experts will have the opportunity to optimize their TensorFlow solution for the cloud TPU environment, using the tools provided by the platform.

 

What Can Google Cloud TPUs Provide?

So, what does the introduction of Google Cloud TPUs mean to the machine learning environment?

Google claims the following benefits of Tensor Processing Units

  • Optimized machine learning: We've already begun to see how machine learning can have a life-changing impact on the way that we live and work. Improvements in NLP (natural language processing) and computer data management will help us to enjoy a stronger relationship with technology. 

  • Faster workloads: The new Cloud TPUs are also designed specifically to accelerate machine learning workloads in the TensorFlow system. Every cloud TPU comes with up to 180 teraflops of performance - ideal for offering the enhanced computing power required for machine-learning models. 

  • ML Supercomputing: Additionally, customers of the Google Cloud Platform will be able to access some of the most impeccable machine learning accelerators on demand - with no up-front requirements for capital. Regardless of what your task requires, you can get all the machine learning support that you need without having to create a new data center. 

  • Easy Access to the Cloud: Thanks to the security and support offered by the TensorFlow system from Google, the product makes it easier for customers to simply try their systems out on the cloud. The APIs are flexible enough to move across TPUs, GPUs, and CPUs with very few coding changes.

  • Access to Google's knowledge: Finally, perhaps most exciting of all, is the fact that Google Cloud Customers will be able to tap into all the incredible data and information that Google uses to develop some of the world's most impressive models for machine learning and AI. The Cloud TPUs have been carefully designed to enhance and support the start-of-the-art systems that Google themselves create.

  

Upgrading Machine Learning with Cloud TPUs

This isn't the first time that Google has experimented with Tensor Processing Units. In fact, they've been using TPUs to run their systems since 2015. However, while the first TPU solution was designed to run machine learning models more efficiently, the models still needed to be trained on separate platforms. Obviously, training a machine learning system is very difficult and time-consuming.

Today, the engineering and research teams at Google have made a huge amount of process into the work that is required to scale and enhance machine learning platforms using hardware that's already available and ready to tap into. After finding that other providers simply didn't have enough solutions to match their machine learning needs, Google designed their very own machine learning system intended to maximize performance and reduce bottlenecks.

The new TPU system available today is the very same one that keeps Google running at breakneck speed, and it comes with all the following features:

  • High performance: Every Cloud TPU comes with 180 teraflops of computing capability, along with 64 gigabytes of amazing high-bandwidth memory.

  • The benefits of the Google Cloud: Machine learning workloads can be run on Cloud TPUs, allowing companies to benefit from Google Cloud Platform's storage, data analytics, and networking solutions.

  • Reference Models: You can solve various object detection and image classification problems on the Cloud TPUs available on Google. What's more, you can even test the language modeling and translation system yourself with the Tensor2Tensor transformer model.

  • Integration: The Google Cloud TPUs and Data & Analytics services are also fully integrated with the remainder of the services that you would expect from the Google Cloud Platform. This means that ever gets complete and unfiltered access to the service line.

  • Connect to Custom Machine Types: Finally, you can connect to the Cloud TPUs from a variety of custom VM types, which helps to balance out a range of different systems, including memory, processing speeds, and storage resources too.

Making the Most of Google Cloud TPUs

Across almost every industry, the machine learning strategy is driving constant innovation. In the manufacturing industry, you have automatic product tracking and industrial supply chain tracking. On top of that, there's the opportunity to verify the quality of parts through cameras that can be installed into the assembly lines themselves. In the retail industry, you can enable the simpler retrieval of products using photos online and offline with tagging apps.

Even in the healthcare industry, machine learning is changing lives by offering better health support based on medical images. A computer can pinpoint the signs of an illness or disease even faster than a human. There are also options for machine learning in the insurance realm when it comes to safety and surveillance, and accessing damage from a car collision.

No matter which sector you're in, if you're ready to explore the true potential of machine learning, then Google has all the tools you need to take your system to the next level. To learn more about TPUs, or find out about the Google Platform, reach out to Coolhead Tech today!