Google officially announced in a blog post that it will open a Tensor Processing Unit (TPU) service to Google Cloud customers for a “limited number.” TPU is a custom artificial intelligence chip designed by Google for machine learning. Opening up the TPU is a good thing for those who choose to run a machine learning model on Google Cloud Platform. However, Google this opening a limited number of TPU, and on time charges, the cost per hour is 6.50 US dollars.
Google launched the TPU project in a low profile in 2016 and launched the second-generation TPU at the annual developer conference in May last year. This update allows Google to enter a more complex stage of deep learning and training.
TPU dedicated to artificial intelligence and machine learning, Google can bring two major benefits:
First, using self-developed chips, Google can reduce its dependence on chip makers such as Intel and NVIDIA in terms of its core computing infrastructure cheaper and more efficient. Having your own hardware allows Google to experiment faster.
Second, the new TPU also brings additional revenue to Google’s cloud computing business, which now generates more than $1 billion in revenue per quarter for Google Cloud Platform (GCP) and Google’s business application GSuite.
Google currently allows companies to rent a single TPU card, and later this year will allow businesses to rent multiple boards and connect a supercomputer network known as the TPU pod.