With this processor Google wants to make the big leap in machine learning

There are still new developments around Google I / O 2016, and this time it is an interesting topic that has become more relevant over time, we refer to artificial intelligence, a field where Google has been very active with developments Capable of beating humans in complex games, and is even capable of nurturing romantic novels to learn how to talk.

But today we know the brain that drives all these developments, a TPU (Tensor Processing Unit) chip, interestingly have been using for a little more than a year, which consists of an ASIC processing unit adapted for TensorFlow, which has been created specifically for machine learning.

TPU, the brain of Google's artificial intelligence


The magic behind TPU is that it is able to prioritize tasks and learn from their mistakes, achieving a large number of operations per minute, but always focusing on those "orders of magnitude", which allows you to have more efficient processes with the same Energy level than other developments based on artificial intelligence.

An important point is that the TPUs are easily scalable, allowing servers to be driven by several TPUs, in order to obtain higher processing speed and to be able to predict the following operations, while learning on the fly, a really impressive job.

The goal of Google with TPU is to lead the learning industry in machines, which will have at our reach new platforms "smart" that require a lesser extent user intervention, platforms that will be available to developers to create applications to operating systems capable of benefiting users and companies. Undoubtedly, one of the most important future-oriented developments that Google has today.

We can say that Google works in a new world and of course, to mark a clear differentiation in terms of optimization and performance, it takes time and development of new solutions that adapt to new issues and needs. TensorFlow is the new open source system for Google's automatic learning, which requires special processors, of course.

Google does not skimp on resources in this field and has bet heavily on this field, it is more, according to the company the new system they are developing would be between ten and thirty times more powerful than for example solutions based on Intel Haswell combined with the NVIDIA K80 . A fundamental aspect in this field is energy efficiency, which is a priority in the industry, in addition to generating less heat, better, something fundamental for the development of data centers. These Google solutions will be for them and nobody else, they will not sell these systems to third companies, neither the physical product, nor the licenses.

This is a world apart, which has thousands of possibilities, solutions and adjustments, with extremely complex systems and require a very advanced knowledge. Personally sometimes it is hard to keep up with the 'conventional' processors and the innovations that are implemented in them, such as to get a head start on all this. Those who like this world can get into this world with a document from which some data can be extracted.
Designed by OddThemes | Distributed by MyBloggerThemes