Tensor Processing Unit (TPU)
The company also introduced a new chipset called Tensor Processing Unit (TPU), a custom application-specific integrated circuit (ASIC) for machine learning that fits in the same footprint of a hard drive.
(TPUs) are custom ASICs Google built specifically for machine learning—and tailored for TensorFlow
, an open-source library for machine learning. Google has been running TPUs inside its data centers for more than a year and has found them to deliver an order of magnitude-better optimized performance per watt for machine learning, Urs Hölzle, Google's senior vice president for technical infrastructure, told reporters.
This is roughly equivalent to fast-forwarding technology about seven years into the future (three generations of Moore's Law). TPUs are tailored to machine-learning applications, allowing the chip to be more tolerant of reduced computational precision, which means it requires fewer transistors per operation. Because of this, TPUs can squeeze more operations per second into the silicon, use more sophisticated and powerful machine-learning models and apply these models more quickly, so users get more intelligent results more rapidly.
A board with a TPU fits into a hard disk-drive slot in Google's data center racks.
"TPUs are used alongside regular CPUs," Hölzle said at an invitation-only press conference. "They're not designed to be the lead processor for anything."
TPUs were the secret sauce for AlphaGo in its recent intelligence challenge match in South Korea.
AlphaGo versus Lee Sedol, or Google DeepMind Challenge Match, was a five-game Go match between South Korean professional Go player Lee Sedol and AlphaGo, a computer Go program developed by Google DeepMind. The match was played in Seoul, South Korea, March 9-15. AlphaGo won all but the fourth game; all games were won by resignation. The match has been compared with the historic chess match between Deep Blue and Garry Kasparov in 1997.
New APIs for Sheets, Slides and Classroom
Google also announced three new APIs for Sheets, Slides and Classroom that enable developers to build feature-rich integrations to help users be more productive using all their business apps.
—Sheets API gives developers programmatic access to nearly all of the features users can add to a spreadsheet;
—Slides API enables developers to export business data from their apps, allowing users to generate and update content and visuals in slide decks; and
— Classroom API adds coursework endpoints that allow developers to sync grades and assignment data between Google Classroom and their applications.
"The new Sheets API, available today, gives developers programmatic access to powerful features in the Sheets web and mobile interfaces, including charts and pivot tables," Sheets Product Manager Tom Holman wrote in his blog.
"For example, developers can use Sheets as part of a rich workflow that pushes data from their app into Sheets and allows users to collaborate on that data before the updated data is pulled back into the original app, removing altogether the need to copy and paste," he wrote.