Google recently announced details of a chip specifically built for machine learning in the form of its Tensor Processing Unit (TPU) that has operated in stealth mode within Google’s data centers for over a year. TPU delivers superior performance for machine learning use cases by processing “more operations per second into the silicon” and is designed to work with TensorFlow, Google’s open source library of machine learning applications. TPU requires fewer transistors per operation and can optimize performance per watt by an order of magnitude for machine learning applications and use cases in what amounts to “fast-forwarding technology about seven years into the future (three generations of Moore’s Law),” according to a Google blog post. TPU currently powers applications such as Google Street View, RankBrain as well as Google Maps and powerfully illustrates not only Google’s commitment to machine learning technologies but also its competitive differentiation as a vendor with the ability to design and operationalize hardware optimized for machine learning applications.
Cloud Computing Today 