Google unveils AI chip: 4 things to know

Google CEO Sundar Pichai discussed a new artificial intelligence chip during his May 17 keynote speech at Google I/O, the company's developer conference in Mountain View, Calif., Wired reports.

Here are four things to know.

1. The new processor chip, called TPU 2.0 or Cloud TPU, enables users to train and execute deep neural networks, a type of machine learning system. Google's original TPU, or tensor processing unit, was only used to execute deep neural networks that had already been trained.

2. Google will not sell the chip, according to Wired. Instead, the company will offer the product on its new cloud service, which it expects to put on the market by the end of 2017. That means any organization on the company's cloud service will be able to build AI software using these processors.

3. This specialized chip will help organizations working with AI. Developers typically train deep neural networks on general chips called graphics processing units, which can take multiple days. However, with a chip designed for training networks, this process may only take a few hours.

4. Even with the specialization, there may be some market barriers. TPU 2.0 is designed for Google's open-source software TensorFlow, although some researchers have already established work on other platforms.

However, Google said it plans to offer the cloud service to researchers for free, as long as they commit to sharing their findings with the public, according to Wired.

More articles on health IT:
Bipartisan bill calls for transparency in federal stockpiling of IT flaws post-WannaCry
UChicago Medicine joins Google effort to reduce readmissions with AI, EHRs
VA signs $19.6M patient scheduling contract with Document Storage Systems

© Copyright ASC COMMUNICATIONS 2019. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.


Top 40 Articles from the Past 6 Months