SAN FRANCISCO (Reuters) – Google unveiled updates to its data center artificial intelligence chips, along with the introduction of a new Arm-based central processor.
Google’s tensor processing units (TPUs) offer an alternative to Nvidia’s advanced AI chips, accessible only through Google’s Cloud Platform. The addition of the Arm-based central processing unit (CPU), dubbed Axion, expands Google’s offerings, with claims of superior performance compared to x86 chips.
“We’re simplifying the transition for customers to Arm,” said Mark Lohmeyer, Google Cloud’s vice president and general manager of compute and machine learning infrastructure. “Axion is designed for seamless integration, allowing users to adopt it without extensive app re-architecting or rewriting.”
Competitors like Amazon and Microsoft have also developed Arm CPUs to differentiate their cloud services. Google, while having previously designed custom chips for various purposes, had not ventured into CPUs until now. While Broadcom collaborated on earlier TPU generations, Google did not disclose any partnership for Axion.
By the Numbers
The new TPU v5p chip is optimized for pods of 8,960 chips, boasting twice the raw performance of its predecessor. Liquid cooling ensures optimal pod performance. Axion offers a 30% performance boost over general-purpose Arm chips and a 50% improvement over current x86 chips from Intel and AMD.
Axion is already in use in select Google services like YouTube Ads in Google Cloud, with plans for wider deployment to the public later this year. TPU v5p is available on Google’s cloud platform starting Tuesday.
Featured Image: Freepik
