Google Cloud Unveils Arm-Based CPUs to Accelerate AI Workloads

At the recent Google Cloud Next 2024 event held in Las Vegas, Google introduced its innovative Arm-based CPU design, named Axion, specifically developed to enhance generative AI workloads in data centers. This cutting-edge CPU line promises superior performance and exceptional energy efficiency.

The Axion chips are reported to deliver performance improvements of up to 50% compared to current x86-based processors. Additionally, Google claims these new CPUs achieve 60% greater energy efficiency than competing hardware solutions. As Amin Vahdat, Google’s Vice President and General Manager for Machine Learning, Systems, and Cloud AI, highlighted in a blog post, “Axion processors combine Google’s silicon expertise with Arm’s highest-performing CPU cores, providing instances with up to 30% better performance than the fastest general-purpose Arm-based instances currently available in the cloud.”

Instead of selling the Axion CPUs directly, Google will offer them for rent via its cloud services, including Google Kubernetes Engine and Dataflow, later this year. Already, Google employs existing Arm-based chips to support various services, such as the Google Earth Engine, the YouTube Ads platform, and BigQuery, Google Cloud's data warehouse service, with intentions to integrate Axion processors soon.

In addition to the Axion announcement, Google Cloud revealed a preview of Arm-based instance support in its Migrate to Virtual Machines service, facilitating the transition of existing applications and workloads to its cloud platform. With this move, Google joins industry giants like Microsoft and Amazon, both of which have recently unveiled their own Arm-based CPUs. Last November, Microsoft introduced Cobalt, tailored for general-purpose cloud workloads, while Amazon launched Graviton4, the latest in its series of custom Arm-based chips.

Rene Haas, CEO of Arm, remarked, “Google’s announcement of the new Axion CPU marks a significant milestone in delivering custom silicon that is optimized for Google’s infrastructure and built on our high-performance Arm Neoverse V2 platform.”

Moreover, Google Cloud presented upgrades to its Tensor Processing Units (TPUs), specialized hardware designed to power accessible enterprise hypercomputers, with the latest version now generally available. Mark Lohmeyer, Vice President and General Manager for Compute and AI Infrastructure, expressed excitement about the introduction of Cloud TPU v5p, describing it as “our most powerful and scalable TPU to date.” This next-generation accelerator is engineered specifically to train large and complex generative AI models.

The Cloud TPU v5p was first showcased last December and has already been integral in training Google’s landmark Gemini large language model, solidifying Google Cloud’s commitment to fostering advanced AI capabilities and solutions in the evolving landscape of technology.