Google launches Axion, the first Arm-based custom data center processor

On Tuesday, Google Cloud joined AWS and Azure in announcing the launch of its first custom Arm processor, called Axion. Google said that based on Arm’s Neoverse 2 design, the performance of its Axion instances is 30% higher than other Arm-based instances from competitors such as AWS and Microsoft. Compared with similar X86-based instances, the performance is 50% higher and the energy efficiency is high. Out 60%.

Google has provided no documentation to support these claims, and,Like us, you probably want to know more about these chips. We asked a lot of questions, but Google politely declined to provide any additional information. No availability date, no pricing, no additional technical data. Those “benchmark” results? The company didn’t even reveal which X86 instance it was comparing Axion to.

“Technical documentation including benchmarks and architectural details will be released later this year,” Google spokesperson Amanda Lam said.

Maybe the chip isn’t ready yet? After all, it took Google a while to announce an Arm chip for the cloud, especially considering that Google has long been building its own in-house TPU AI chips and, more recently, custom Arm-based mobile chips for its Pixel phones. AWS launched Graviton chips as early as 2018.

To be fair, though, Microsoft only released its Cobalt Arm chips late last year, and those chips aren’t yet available to customers. But Microsoft Azure has provided instances of Ampere-based Arm servers since 2022.

In a press release ahead of Tuesday’s announcement, Google emphasized that because Axion is built on an open foundation, Google Cloud customers will be able to bring existing Arm workloads to Google Cloud without any modifications. . This is indeed not surprising. Any other move would be a very foolish move for Google Cloud.

“We recently contributed to SystemReady virtual environments, Arm’s hardware and firmware interoperability standard that ensures common operating systems and software packages can run seamlessly on ARM-based systems,” Google Cloud Compute and AI/ Mark Lohmeyer, Vice President of ML Infrastructure, explained. “Through this collaboration, we are accessing a broad ecosystem of cloud customers who are already deploying ARM-based workloads across hundreds of ISVs and open source projects.”

There will be more later this year.