Google was one of the first hyperscalers to build custom chips for its services, first using tensor processing units (TPUs) for its artificial intelligence initiatives and then video transcoding units (VCUs) for its YouTube service. But unlike its peers in the industry, the company has been slow to adopt custom CPU designs, preferring to stick with off-the-shelf chips for its main CPUs. That finally changed as Google announced the development of Axion, its own in-house data center CPU.
Google’s Axion processor is based on the Arm Neoverse V2 (Arm v9) platform, which is Arm’s latest generation of high-performance server CPU design and is already used in other chips such as NVIDIA’s Grace. Within Google, Axion targets a variety of workloads, including web and application servers, data analytics, microservices, and AI training. Google claims that Axion processors offer 50% better performance and 60% higher energy efficiency than current x86-based processors, and 30% better performance than Arm-based CPUs used in data centers. While the mysterious cloud side of Google’s business is increasingly common, at least for now the company hasn’t specified which processors they’re comparing the Axion to in these metrics.
While Google hasn’t revealed the core count or full specs of its Axion CPU, the company has revealed that it’s building its own secret sauce into the chip in the form of the company’s Titanium dedicated microcontroller. These microcontrollers are designed to handle basic operations such as networking and security, as well as offload storage I/O processing to the Hyperdisk block storage service. Due to this offloading, almost all CPU core resources should be available for real workloads. As for the chip’s memory subsystem, Axion uses traditional dual-rank DDR5 memory modules.
Rene Haas, CEO of Arm, said: “Google’s new Axion CPU marks an important milestone in delivering custom chips that are optimized for Google’s infrastructure and built on our on the high-performance Arm Neoverse V2 platform.” “Decades of ecosystem investment, coupled with Google’s continued innovation and open source software contributions, ensure the best experience for customers’ most important workloads wherever they run Arm.”
Google has previously deployed Arm-based processors for its own services, including BigTable, Spanner, BigQuery and YouTube Ads, and is preparing to offer instances of Armv9-based Axion CPUs to its customers that can use software developed for the Arm architecture .
Source: Google, Wall Street Journal