Google’s machine learning chip, Apache Software Foundation’s new executive director,…

There have been plenty of announcements coming out of Google I/O this week, and yesterday, Google said that one of its projects that was created years ago is helping it accomplish its own custom accelerators for machine learning applications. The result of that project is called a Tensor Processing Unit (TPU), which is a custom ASIC that was built by Google just for machine learning. TPUs run inside Google data centers and have been for more than a year, according to a Google blog post by Norm Jouppi, a hardware engineer at Google. The company found that TPUs deliver better-optimized performance per watt for machine learning. The chip requires fewer transistors per operation, meaning Google can squeeze more operations per second into the silicon, according to Jouppi. A board with…


Link to Full Article: Google’s machine learning chip, Apache Software Foundation’s new executive director,…