Amazon deploys ARM-based Graviton processor
The AWS Graviton processor is based on 64bit Neoverse cores, Amazon said without providing further detail of the hardware. ARM’s Drew Henry senior vice president and general manager, Infrastructure Line of Business, said that Graviton is based on the Cosmos 16nm processor platform (see Reports: ARM discloses roadmap performance in Taiwan).
According to reports the Graviton is based on the Cortex-A72 64bit core operated at clock frequencies up to 2.3GHz. What is not clear is how many cores are included per physical chip.
The Cortex-A72 is a relatively mature design that would have typically been designed to target 16nm manufacturing processes. Servers based on Intel and AMD processors are expected to be faster in terms of performance but home-designed ARM SoCs may provide Amazon with a performance-per-dollar advantage quoted by Amazon. And there is the possibility of an using the Cortex-A76 in 10nm.
The Graviton design has come out of Israeli design house Annapurna Labs. Annapurna, founded in 2011, was acquired before it had produced any products, by Amazon in January 2015 for a sum reported to be in the range $350 million to $370 million. In 2016 it had its own line of ARM-based processors (see Amazon now sells its own ARM chips).
Next: ARM in data center
Graviton is Amazon’s first use of in-house designed processors in its cloud platform with Annapurna’s previous range being used in home gateway and Wi-Fi router products from Netgear, QNAP Systems and Synology.
“I’ve been interested in ARM server processors for more than a decade so its super exciting to see the AWS Graviton finally public, it’s going to be exciting to see what customers do with the new A1 instances, and I’m already looking forward to follow-on offerings as we continue to listen to customers and enhance the world’s broadest cloud computing instance selection,” said James Hamilton, vice president and distinguished engineer, in a blog. “There is much more coming both in ARM-based instance offerings and, more broadly, across the entire of the Amazon EC2 family.
Related links and articles: