Graviton progress: 50% of new AWS instances run on Amazon custom silicon

Half of all instances spun up on Amazon Web Services over the past two years used Graviton, Amazon’s custom silicon built on Arm processor technology, rather than x86 processors. In addition, the homegrown Graviton processor powered more than 250,000 CPUs during Amazon Prime Day, showing its efficiency in handling demand, according to a presentation at the recent AWS: reInvent conference by Dave Brown, vice president of AWS compute & networking services.

So, while Graviton may not be for sale to anyone outside of Amazon, AWS is getting a major return on its investment.

For starters, the AWS disclosure confirms that hyperscalers are developing more and more of their own custom silicon, according to Patrick Moorhead, CEO of Moor Insights & Strategies.

“You always have to ask, well, how much of it are they using? Is this some sort of a, ‘hey, we’re going to kick the tires on this thing?’ But when somebody shows up and says that 50% of their new capacity over the last couple years has been their own custom CPU, I think that that is a huge statement that says you can do this at scale,” Moorhead said.

AWS claims nearly every one of its customers is using Graviton to some degree, according to Moorhead, who attended Brown’s keynote. “What that indicates is that it’s pervasive across their customer set and not necessarily concentrated [with one group]. So, it’s more of a trend as opposed to a one-off,” he said.

The primary appeal of Graviton is savings. Graviton instances cost up to 20% less and use up to 60% less energy than comparable EC2 instances on x86, and they are up to 40% more cost-effective than x86.

Graviton’s success could impact Ampere, which makes Arm-based processors for both the enterprise and the cloud. Ampere is challenged in that all the hyperscale cloud providers – AWS, Microsoft, Google, and Meta – are making their own custom chips rather than using a third-party processor.

But the challenge of getting Arm into enterprise data centers lies in all the legacy code. There are a lot of homegrown and packaged applications written for x86 processors that are not available for Arm, which will lead some enterprises to stick with x86 infrastructure.

Since its introduction in 2018, Graviton has gone through four generations, which is a considerable pace for a company with no silicon design experience.

In July 2024, AWS announced the launch of its fourth-generation Graviton CPU, touting its energy efficiency and high performance for cloud workloads. Graviton4 offers a significant performance upgrade over Graviton3, with 30% better computing power, 50% more cores and 75% more memory bandwidth. The new Graviton4 instances, called R8g, support up to 8GB of memory per virtual processor and up to 192 processors.

Other hyperscalers have also been able to jumpstart Arm projects.

Amazon, Microsoft, Google, and Nvidia have quickly brought their respective enterprise efforts to market thanks to a custom compute subsystem from Arm called Arm CSS, which helps partners by providing extra subsystems like memory and interconnections. The hyperscalers then differentiate their designs from the competition by using different networking and security protocols, among other things.

Source:: Network World