Google Says Its Machine Learning Chip Leaves CPUs, GPUs inside the Dust

By | April 6, 2017

Without its custom-designed chip for machine learning, Google would possibly want wished to assemble many extra information facilities to assist the intensive computing calls for of providers akin to Image Search, Google Photos and more.

Yesterday Google revealed new efficiency particulars about its Tensor Processing Unit (TPU) all by means of a National Academy of Engineering presentation on the Computer History Museum in Mountain View, Calif. It said the TPU has proven in exams that it might presumably outperform CPUs and GPUs in information centers, enabling sooner and extra energy-efficient computing for superior demands.

As it utilized machine studying capabilities to extra of its merchandise and functions over the previous a quantity of years, Google said it realized it wished to supercharge its hardware as effectively as to its software. It launched a “stealthy” mission to develop a custom-made accelerator, and final May revealed that it had been effectively working Tensor Processing Units in its information facilities for greater than a year.

Faster Chip or Twice as Many Data Centers

“The want for TPUs actually emerged about six years ago, as quickly as we started using computationally costly deep studying fashions in more and more places all by means of our products,” Norm Jouppi, a distinguished hardware engineer at Google, wrote yesterday on Google’s Cloud Platform Blog. “The computational expense of using these fashions had us worried.”

According to Jouppi, the agency calculated that Google prospects using voice look for simply three minutes a day would have overwhelmed current information center processors working deep neural nets for speech recognition. Keeping up would have required Google to double its current quantity of information centers, he said.

Google at present operates 15 foremost information facilities throughout the globe, with eight positioned inside the us and the relaxation scattered throughout Europe, Asia and South America.

The agency launched its high-priority TPU development mission with the purpose of boosting computation…

Please check this great service at: http://www.test-net.org/services/proxy-checker/ or visit FREE SERVICES menu

[Total: 0    Average: 0/5]