Jump to Content
AI & Machine Learning

Introducing TensorFlow Enterprise: Supported, scalable, and seamless TensorFlow in the cloud

October 30, 2019
Craig Wiley

Director of Product Management, Cloud AI and Industry Solutions

If you’re developing AI, you’re likely using TensorFlow. Open-sourced by Google in 2015, it’s grown to be one of the most popular machine learning frameworks in the world. But its enterprise users have higher demands and expectations for running supported, scalable, and seamless ML workloads. 

Introducing TensorFlow Enterprise

To address the needs of AI-enabled businesses, we are introducing TensorFlow Enterprise. TensorFlow Enterprise incorporates: 

  • Enterprise-grade support 

  • Cloud scale performance

  • Managed services

Together, these services and products can accelerate your software development and ensure the reliability of your AI applications.

https://storage.googleapis.com/gweb-cloudblog-publish/images/Google_Cloud_TensorFlow_Enterprise_1.max-1100x1100.jpg

Enterprise-grade support

The pace of AI and software versions is evolving rapidly, but many customers have told us they are heavily invested in a previous version of TensorFlow. That is why TensorFlow Enterprise includes long-term version support. For certain versions of TensorFlow, we will provide security patches and select bug fixes for up to 3 years. These versions will be supported on Google Cloud, and all patches and bug fixes will be available in the mainline TensorFlow code repository.

For customers where AI is their business, we have an additional offering. TensorFlow Enterprise offers a white-glove service to help these cutting-edge customers tackle their biggest AI challenges. It includes engineer-to-engineer assistance from both Google Cloud and TensorFlow teams at Google.

Cloud scale performance
The skills to start a business are often not the same as those needed to scale and grow a business, and the same is true for AI development. Many models begin as an idea and a single-node on-prem, and scaling to the performance potential of the cloud can be daunting. 

Google Cloud offers a range of compute options for training and deploying models. TensorFlow Enterprise includes Deep Learning VMs (GA) and Deep Learning Containers (Beta), which make it simple to get started and scale. Both products are compatibility tested and performance optimized for our wide range of NVIDIA GPUs and our custom-designed AI processor, the Cloud TPU. 

One such performance improvement is in data throughput. In order to feed these accelerators, TensorFlow Enterprise optimizations have increased data reading times by up to 3x, drastically accelerating workloads. 

Unity Technologies uses this speed and performance to drive revenue for developers on their platform. 

"Unity’s Monetization products reach more than 3 billion devices worldwide. Game developers rely on that mix of scale and products to drive new installs, revenue and player engagement. Through the power of Google Cloud’s TensorFlow Enterprise, we can quickly test, build and scale our Machine Learning models at massive scale, allowing us to serve up the most relevant ads and drive revenue for game developers.” -Jeff Collins, Unity Technologies, VP Engineering, Monetization

Across Google Cloud
TensorFlow Enterprise offers the same optimized experience and enterprise-grade features across Google Cloud managed services, like Kubernetes Engine and AI Platform. Whatever stage of development you’re in, from development to deployment, Google Cloud offers an end-to-end workflow on TensorFlow. 

The Best Place to Run TensorFlow

Because Google created and open-sourced TensorFlow, Google Cloud is uniquely positioned to offer support and insights directly from the TensorFlow team itself. Combined with our deep expertise in AI and machine learning, this makes TensorFlow Enterprise the best way to run TensorFlow.

For example, GM Cruise uses TensorFlow Enterprise to accelerate their autonomous driving capabilities. 

"Developing and deploying self-driving vehicles at massive scale is the engineering challenge of our generation. The unique collaboration across both Google Cloud and TensorFlow teams was critical in helping Cruise solve our most pressing scaling issues. As a result, we increased the accuracy of our models while reducing training times from over four days to less than one. TensorFlow Enterprise enabled us to build better models, faster.”  -Hussein Mehanna, Head of AI/ML, GM Cruise

Get Started with TensorFlow Enterprise 

TensorFlow Enterprise is available to you today. Learn more at the TensorFlow Enterprise web page, try out a hands-on codelab, and get started immediately with how-to-guides

For more information, please see our technical blog post on Tensorflow Enterprise. If you qualify and would like to be considered for the white-glove service of TensorFlow Enterprise, you can apply here.

Posted in