Google (NASDAQ: GOOGL) has revealed its Tensor Processing Unit, named after TensorFlow, a software that allows the chip to undertake machine learning.
The Tensor Processing Unit or TPU is an accelerator chip, according to Google’s Norm Jouppi. The TPU is able to put more processing speed towards processes that involve voice recognition, photo identification and analysis of big data. In addition to any task that requires the processor to learn through analysis and raw input.
The TPU can also be regarded as an ASIC, an application-specific integrated circuit. The TensorFlow software is open-source, meaning that anyone can program the software engine to fit its needs. Making its users able to develop more advanced applications as well.
More machine learning
Facebook, Twitter and most tech-giants have taken the subject of machine learning very seriously. Most of these tasks are developed through the use of standard graphics processing units developed by Radeon or Nvidia.
Using these GPUs for machine learning has proved to be inefficient. Thus, Google has given an answer by letting the world know about the plans behind the novel TPU. According to the Alphabet-owned company, the TPU is able to provide a better ‘performance per watt’ for machine learning when compared to other kinds of hardware.
The statement revealed that Google has ceased to use GPUs over the past few years. As they have displayed their ability of developing their own chips, Google has become a direct threat for chip-manufacturing companies. Which includes Nvidia and Intel at the very top of the list. The TPUs have been under Google’s sleeve for nearly a year. Yet it is worth noting that its pre-implementation period lasted only 22 days, according to Google.
Have a question about Machine Learning? Maybe @NatAndLo can get it answered in their next video episode on ML https://t.co/UwVFjcuOce
— Google Research (@googleresearch) April 29, 2016
Google is set to innovate with Machine Learning
Some of the applications currently using Google’s TPU are Street View and RankBrain. Thanks to the TPUs, there is a much better precision when downloading personalized maps on Street View. RankBrain on the other hand, allows users to get better results on Google Search in all of their apps and services.
It seems that Google wants to become the spearhead in the machine learning industry. And intends to do it so by creating specialized software and hardware. The TPU is the perfect turning point as it is a very specific piece of hardware. A little piece of hardware with a very specific goal, improving the efficiency of machine learning.
The company is also aiming towards massively implementing machine learning on the Google cloud. The easiest way for an user to make use of the TPUs in Google’s servers is by using Google voice recognition on a search query. Thousands of TPUs are in use right now in Google, it is a matter of time until information technology gargantuans implement them on their own hardware.
Built for machine learning. Tailored for #TensorFlow. Learn how our custom TPU chips work: https://t.co/nnnZH5OHFD pic.twitter.com/8u8jkLuZ71
— GoogleCloudPlatform (@googlecloud) May 18, 2016
Source: Google Blog