If you’ve ever opened up your mobile phone, either by design or by unfortunate accident, you’ll likely find computing chips that are based on what is known as the Arm architecture. Arm-based chips are extremely powerful and extremely power-efficient—they won’t kill your battery, in other words—which is why they are perfect for phones. Those advantages are also why, in recent years, Arm chips have been used in more than just phones. They’re deployed in all kinds of computing environments—except the cloud.
The reason? Turns out it’s really hard. Others had tried before and failed. Arm is just the starting point, the engineering team at Annapurna Labs began with Arm’s customizable nature and then landed on their own refined chip design and process.
The result was Graviton.
AWS Graviton4Launched in 2018, the custom AWS-engineered data center chip was the first of its kind to be deployed at scale by a major cloud provider. Graviton brought the same efficiency to AWS data centers, and all the power required by AWS customers.
How powerful? Graviton4 offers four times the performance of Graviton1. How efficient? Graviton3 uses 60 percent less energy for the same performance as comparable Amazon EC2 instances (where the compute happens in a data center), and Graviton4 is even more energy efficient. How many transistors are in the chip? Glad you asked: 73 billion transistors, which in compute terms is the equivalent of a whole, whole, lot.
SAP, Epic Games, and SmugMug are among the AWS customers who are already benefiting from using Graviton4-based instances. Epic, operator of Fortnite, one of the world’s largest games with over 350 million accounts, is using AWS Graviton4 instances to deliver gaming the way Fortnite players expect, as fast as possible and as reliably as possible.
AWS has been driving the industry embrace of processors designed explicitly for cloud workloads, and has been at the forefront of enabling their broad use in cloud applications as well as working with partners to make them accessible to more and more customers. With Graviton, AWS was, and is, at the forefront of this custom chip evolution. But the revolution in custom silicon for the cloud doesn’t end there. With the latest generation of AWS chips engineered for AI, Trainium and Inferentia, AWS is extending the chip development environment—the combination of hardware and supporting software—and the success that began five years ago with Graviton.
Check out some highlights of the four generations of Graviton chips.
01 / 06
