< Until June to wait for CW lead art > hidden potential value chess Bishop 7 How to get the most from your dealer Strategy for several years, in new ways, such as consolidating a huge change you Read now. It calls the chip processing unit or TPU, it uses machine learning program named after the TensorFlow software. In a blog post, Google engineer norm Jouppi refers to it as the accelerator chip, which means that it has accelerated a specific task.
At the Google I/O Conference, summer on Wednesday, Chief Executive, TPU provides orders of magnitude better performance per watt than previous chips machine learning task. It does not replace the Cpu and GPU, but it can speed up the learning process, without consuming more and more energy.
As machine learning becomes more widely used in all types of applications, from voice recognition to language translation and data analytics, having a chip that speeds those workloads is essential to maintaining the pace of advancements.
And as Moore's Law
slows down, reducing the gains from each new generation of processor, using accelerators for key tasks becomes even more important. Google says its TPU provides the equivalent gains to moving Moore's Law forward by three generations, or about seven years.
TPU in the Google cloud, including powered RankBrain search result sorting system and Google's speech recognition services, are used in a production environment. When the developers pay to use Google voice recognition service, they use the GPU.
TPU can complement machine learning processes, but there is still need for Cpu and GPU features, Urs Hall, technology infrastructure, Google Senior Vice President said at a press conference at the i/o Conference.
TPU Google began to develop about two years ago, he said.
Currently, Google has thousands of chips used. They can be in the same slot for your hard disk in a Google data center rack, which means that companies can easily deploy more people if it needs to adapt. POOPP['
Right now, though, Hölzle says that they don't need to have a TPU in every rack just yet.
If there's one thing that Google likely won't do, it's sell TPUs as standalone hardware. Asked about that possibility, Google enterprise chief Diane Greene said that the company isn't planning to sell them for other companies to use.
Part of that has to do with the way application development is heading -- developers are building more and more applications in the cloud only and don't want to worry about managing hardware configurations, maintenance, and updates.
Another possible reason is that Google simply doesn't want to give its rivals access to the chips, which it likely spent a lot of time and money developing.
What we still don't know what TPU are better used to do. Analyst Patrick · Moorehead said he expects the chips will be used in inference, machine learning part of the operation, do not need as much flexibility as possible.
Now, this is all Google said. We still don't know which one is for Google to build silicon chip manufacturer. Holzle says the company will reveal more about the chip in a document, released this fall.
