During the Google I / O, the company presented the TPU 3.0, intended for Artificial Intelligence, which has a liquid cooling system, for the first time.
Artificial Intelligence has become a very lucrative industry and one in which none of the large companies in the market wants to lose ground. One of the most cutting-edge is Google, which not only develops tools for AI, but also hardware. Google in 2015 introduced for the first time the Tensor Processing Unit or TPU. These systems are used in Data Centers and have been developed to work specifically under TensorFlow framework, a symbolic mathematical library that is used for AI in Machine Learning solutions, such as artificial neural networks.
During the Google I / O they presented the TPU 3.0, a new processing system for Machine Learning that consists of great power. These drives are eight times more powerful than the predecessor drives from 2015 and are made up of ASICs based on 64GB of HBM2 memory with a bandwidth of 2400GB / s.
'These chips are so powerful that, for the first time, we must introduce liquid cooling in our data centers. We have put these chips in the form of giant capsules and each one of them has 8 times more power than last year's, over 100 Petaflops, and this will help us develop better, bigger and more precise models that help to drive bigger problems. ' Sundar Pichai, CEO of Google, has highlighted.
Nobody wants to be left behind in the race of Artificial Intelligence and Google is one of the most advanced. Above all, this system stands out for being based on a liquid cooling system, which, as Pichai highlights, is the first time that it has been used for this type of system, used in this case, due to the great power it offers.
Source: TT
