Google unveils new AI chip and supercomputer

Google CEO, Sundar Pichai, announced a new computer processor at the company’s annual developer conference. The chip is designed to perform machine learning and show just how quickly artificial intelligence is taking over. It also displays how the company plans to establish itself as a software and hardware developer.

What’s most notable about the new processor is that it operates at an extremely fast speed and can be trained efficiently. Deemed the Cloud Tensor Processing Unit, the chip was created and named after Google’s open-source TensorFlow machine-learning system. Since training is a huge part of machine learning, the calculations required for such a considerable model are so complex that it may take days or weeks to complete.

Additionally, Google unveiled machine-learning computers, called Cloud TPU pods. The computers are based on a string of Cloud TPUs wired together with high-speed data connections. Pichai also noted that Google is creating the TensorFlow Research Cloud, a space with thousands of TPUs accessible via the internet.

“These TPUs deliver a staggering 128 teraflops, and are built for just the kind of number crunching that drives machine learning today,” said Fei-Fei Li, chief scientist at Google Cloud and the director of Stanford’s AI Lab.

One-thousand Cloud TBU systems will be made available to AI researchers on board with openly sharing details of their work.

During the conference, Pichai also unveiled several AI actions. An effort is being made to create algorithms that can learn how to do tedious work, such as that involved with other machine-learning algorithms. Google is also developing AI tools for medical image analysis, genomic analysis, and molecule discovery.

The move to focus on AI hardware and cloud services is due, in part, to Google’s attempt to speed up its own operations. The company itself uses TensorFlow to power search, speech recognition, translation, and image processing. It was also used to program AlphaGo, Google’s robotic Go player.

Via Technology Review