Advertising giant Google has unveiled a custom processor developed to speed up its TensorFlow machine learning platform: the Tensor Processing Unit, or TPU.
Since its public release back in November
, there has been considerable interest in using Google's machine learning platform for a variety of tasks - not least inside Google itself. The company uses the platform for everything from natural-language processing to image recognition, and now it has a way to make its operation considerably faster: the Tensor Processing Unit.
'We started a stealthy project at Google several years ago to see what we could accomplish with our own custom accelerators for machine learning applications. The result is called a Tensor Processing Unit (TPU), a custom ASIC we built specifically for machine learning — and tailored for TensorFlow,
' explained Google's Norm Jouppi in the announcement
. 'We’ve been running TPUs inside our data centers for more than a year, and have found them to deliver an order of magnitude better-optimised performance per watt for machine learning. This is roughly equivalent to fast-forwarding technology about seven years into the future (three generations of Moore’s Law).
'TPU is 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, we 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 our data center racks.
The TPU is far from a lab experiment, too: impressively, Google went from first tested silicon to running applications within its data centres in just 22 days and currently accelerates products from Street View and RankBrain to the machine intelligence which bested Go champion Lee Sedol recently. Thus far, though, the company is silent on whether it plans to sell TPU units to outside customers or simply make it a unique offering for its own cloud platform instead.