Google today announced its next-generation Tensor Processing Units (TPUs), which aim to make Google Cloud Platform (GCP) “the best cloud for machine learning,” CEO Sundar Pichai said during his keynote at Google I/O.
These Cloud TPUs, available to companies via Google Compute Engine, support larger amounts of machine learning compared to their first-generation counterparts launched at Google’s annual developers conference last year.
Each Cloud TPU device delivers up to 180 teraflops of floating-point performance, Pichai said. They are designed for data centers and can be stacked to build “machine learning supercomputers” called TPU pods. A TPU pod contains 64 TPUs and provides up to 11.5 petaflops.
Google will also make 1,000 Cloud TPUs available for free to machine learning researchers via the TensorFlow Research Cloud.
AI, Cloud Driving Google
During his keynote, delivered to more than 7,000 people at Shoreline Amphitheatre in Mountain View, California, Pichai said Google’s investment in machine learning reflects an overall shift in computing.
“This shift from a mobile-first to an AI-first world, and we are driving it forward across all of our products and platforms,” Pichai said, echoing his comments to investors on the company’s latest earnings call last month. On the call, Pichai said machine learning and artificial intelligence (AI) are fueling innovation across the entire company.
Also this week Google jumped into the Internet of Things (IoT) business and made its enterprise cloud database service generally available. The company launched a beta version of Cloud Spanner as a service on Google Cloud Platform in February.
On May 16, Google product manager Dominic Preuss said in a blog post that customers including Snap are already using the database service. “Cloud Spanner is already delivering long-term value for our customers with mission-critical applications in the cloud, including customer authentication systems, business-transaction and inventory-management systems, and high-volume media systems that require low latency and high throughput,” he wrote.
The same day, GCP launched Google Cloud IoT Core, a managed service to help enterprises capitalize on the data from all the sensors connected to their industrial devices.