Evolving cloud workloads are going to require evolving hardware strategies to anticipate those future demands, and Amazon Web Services continues to lean on its 2015 acquisition of Annapurna Labs to create custom chips that will give customers a wide range of performance options.
AWS vice president of infrastructure and customer support Peter DeSantis closed out the 2018 GeekWire Cloud Tech Summit with a discussion of the custom chips that AWS is developing that deliver the specific type of horsepower needed to address challenging workloads that are moving onto the cloud, such as drug discovery, oil and gas exploration, and artificial intelligence.
“This ability to go all the way — to putting your infrastructure into custom design — has truly allowed us to take a bunch of customer requirements that excite us and invest in really large efforts to deliver this value,” DeSantis said.
During the early days of cloud computing, customers were content to have familiar standard-issue hardware at their fingertips through cloud providers like AWS and Microsoft Azure. In essence, they were building software the same way they always had; they were just running it on general-purpose Intel processors and virtual machines on servers managed by someone else, rather than setting up racks and racks of their own equipment.
There’s still an awful lot of customers who are still satisfied with that arrangement in 2018, but there are also growing numbers of customers that want to put more challenging workloads on the cloud, and that requires very specialized chips that go beyond the bounds of general-purpose processors. Artificial intelligence research was one of the first disciplines that caught the eye of cloud providers, as they scrambled to offer Nvidia’s GPUs — graphics processing units designed to be really good at one specific type of workload — to their customers.
DeSantis discussed how AWS is expanding its “Nitro” chip program, which it first discussed back in November at AWS re:Invent 2017.
“The chips we’re primarily talking about here, they’ve really just gotten EC2 out of the customer’s way,” DeSantis said, referring to how specialized processors allow customers to avoid creating a lot of custom software to extract the performance they need from AWS’s general-purpose cloud computing instances.
DeSantis also hinted at a few new capabilities coming to this program, including a new hypervisor — the software that allows for the creation of virtual machines — and new security features.
AWS is hardly alone in pursuing custom silicon for certain cloud computing features. Google’s TPU processors are in their their second generation, and Microsoft’s Project Brainwave aims to deliver the same type of performance to Azure customers.