Google brings Arm-based Axion processors to train AI models: Details here

Google Axion processor is already powering core services like Google Earth Engine and YouTube Ads platform and more

Google has announced its own custom Arm-based CPU, Axion, specifically built for data centres. The new Axion processors are capable of handling general-purpose workloads, including CPU-based artificial intelligence (AI) training and media processing.

Google said in a press note that Axion processors deliver up to 30 per cent better performance than the currently available fastest general-purpose Arm-based processors. Additionally, Google Axion processors offer up to 50 per cent improvement in performance and up to 60 per cent better energy-efficiency than comparable x86-based solutions, according to the company.

Multiple Google services like Google Earth Engine and YouTube Ads platform are already powered by the Axion processors and the company plans to make it available to Cloud customers later this year. Google said that customers would be able to use Axion processors in many Google Cloud services, including Google Compute Engine, Google Kubernetes Engine, Dataproc, Dataflow and Cloud Batch.

Google on X (formerly Twitter)

Axion processors are not the first-in-line Google silicon. The American technology giant has already released five generations of Tensor Processing Units (TPU) since 2015, specifically made to accelerate AI and machine learning applications. Google also has its own line of Tensor system-on-chips (SoC) that power Pixel smartphones and other devices.

 

Last year, Microsoft introduced its own Arm-based custom AI chips for Azure Cloud services, named Azure Maia 100 and Azure Cobalt 100 CPU. The Mia 100 chip is designed to run large language models and can be used to train AI models, while Cobalt, a custom Arm-based CPU, is designed to handle general computing workload.