Google introduces Axion, the company's first custom Arm data center processor. Axion is based on Arm Neoverse V2 architecture and is said to be 'thirty percent faster than current Arm competitors' chips. The chip will be available to Google Cloud customers later this year.
Google writes in a blog post that Axion is based on the Arm Neoverse V2 architecture. That is a 2022 Arm platform that is used for data center and server chips and is designed for cloud applications, high performance computing and machine learning. The chip also has a system that Google calls Titanium. That includes microcontrollers for specific tasks, such as networking and security. The CPU itself must therefore have more computing capacity for general computing tasks.
The company also shares few concrete specifications about the chips. For example, it is not known how many Neoverse V2 cores the processor has. Google does share some performance claims based on its own testing. The tech company claims that Axion is up to 30 percent faster than “the fastest general-purpose Arm instances currently available.” The company also says that Axion performs up to 50 percent better and is 60 percent more efficient than “comparable x86-based instances.” However, the company does not share what data those figures are based on. Google also does not say exactly which competing chips were used for those comparisons.
Google wants to use Axion in its own cloud services soon. Customers should soon also be able to use the chip via Google Cloud applications such as Compute Engine, Google Kubernetes Engine, Dataproc, Dataflow and Cloudbarch.
With Axion, Google is following the example of other cloud giants, who have been working on self-designed solutions for some time. Arm processors. For example, Amazon presented the Graviton4 SoC with 96 Arm Neoverse V2 cores in November 2023. Microsoft also unveiled its first custom Arm chip last year, called Cobalt 100. That chip should eventually become available to Azure customers. Axion is Google's first self-designed Arm chip for data centers, although the manufacturer has been developing TPU accelerators for years. These are used for specific AI calculation tasks.
:strip_exif() /i/2006622330.jpeg?f=imagenormal)
The Google Axion processor. Source: Google Cloud