Google has revealed that it has made its own chip dedicated to AI and machine learning.
We don’t really think of Google as a component manufacturer, but the truth is that the company is willing to get its hands dirty wherever it deems it necessary to do so.
One of those areas, it seems, is artificial intelligence, where the internet giant is something of a pioneer. It seems current CPU technology simply wasn’t providing optimal results, so Google designed its own chip. As Google I/O gets into gear, Google has found time for an interesting aside.
Several years ago, reveals a fresh blog post, Google started a stealthy project to see how far Google could push machine learning from a hardware standpoint. The result was the Tensor Processing Unit (TPU).
This custom chip is said to offer “an order of magnitude better-optimised performance per watt for machine learning” than current off-the-shelve chips. Or, to put it another way, “This is roughly equivalent to fast-forwarding technology about seven years into the future (three generations of Moore’s Law).”
Though Google is only talking about this now, it has actually been running its TPUs inside its data centres for more than a year. See, we said that it was stealthy.
Related: How to install Android N right now
In fact, you’ll have seen the result of its added oomph in the news this year, as the AlphaGo AI that beat Go world champion Lee Sedol was powered by Google’s TPUs.
You’ll also have benefitted directly from the performance of the chip, as Google has used its TPU technology to boost RankBrain (for improved search results) and Street View (for improved maps and navigation).
Next, take a look at our Android N hands-on:
And you thought Google was just a software company.