Amazon Web Services has made it possible to use its Arm-powered, home-built Graviton2 processors with its Lambda serverless functions.
On Wednesday, the server tenant announced that Gravitons were an option alongside its x86 offerings from Intel and AMD.
It’s no surprise that AWS calculates its own silicon costs on price – the company says Graviton2 can deliver “up to 19% better performance at 20% lower cost.”
But wait: there is more! AWS asserted that “workloads that use multithreading and multiprocessing, or perform a lot of I / O operations, may experience lower execution time and, therefore, even lower costs.”
AWS finds that it is not difficult to test this claim if your functions do not have binary dependencies – a fortunate lack that will often be the case for functions written with Node.js and Python, or functions compiled in Java bytecode.
Under these circumstances, the colossus of the cloud suggests that “changing the architecture of a Lambda function is like flipping a switch.” Or, to be more precise, changing your run settings to select
arm64 in the place of
Serverless functions are not always lightweight which can get away with modest compute resources. Indeed, in December 2020, AWS tripled the memory it was willing to put behind Lambda functions, raising the cap to 10 GB of RAM and allowing half a dozen vCPUs. This specification remains an option with Graviton2, so it’s not as if AWS suggests that you run functions on just one of Graviton2’s 64 cores.
AWS also doesn’t restrict Graviton2 to tasks that at first glance seem best suited for a Xeon or EPYC: the company allows it to handle heavy workloads, including databases requiring 64 cores and a terabyte of RAM.
Lambda is the 25th service AWS runs on silicon.
Amazon.com’s side business isn’t the only one that uses Arm-powered servers to deliver cheaper cloud services – Oracle has done the same. Among the hyperscalers, Microsoft has also shown signs of being arm-curious. Â®