LAS VEGAS – Anyone who has built a data center or cloud service provider over the last 15 years knows that there haven’t been a lot of options when putting together state-of-the-art hardware. Amazon Web Services’ Peter DeSantis thinks that is about to change.
One of the biggest announcements this week at re:Invent 2018 was the Graviton chip, a new custom-designed server processor based on the Arm architecture that AWS is immediately serving up to its cloud customers. Graviton won’t necessarily kick Intel out of the millions of servers humming in AWS data centers around the world, but it does represent a new era of hardware design in which AWS will be able to move fairly quickly based on customer feedback to introduce new chips designed for the cloud-computing era, during which many of the standard practices of the past have started to look outdated.
“I think we’re going to see innovation again and a lot of innovation,” DeSantis said in an interview with GeekWire. “That’s what I’m truly excited about and I don’t yet know what that’s going to lead to. Nobody does.”
Graviton will likely be the first server processor designed around the Arm architecture to make a significant impact on enterprise computing. Processor cores designed by Arm are ubiquitous in mobile computing, used by companies like Apple and Samsung as the foundation for their own chip designs, but have never established a foothold in the server world.
Arm cores are very power efficient, which makes them ideal for mobile computing and also appeals to companies running massive arrays of data centers like AWS, given their monthly electricity bills. However, they can’t yet match the raw performance of chips from Intel and AMD.
DeSantis envisions that its Graviton A1 processor will be used by companies for web serving or microservices, with customers whose applications need as much performance as they can get sticking with other cloud instances from AWS powered by Intel or AMD. But the processor could be attractive to anyone running applications that depend on the classic “scale-out” data center architecture, in which lots of relatively cheap servers are harnessed together in a fashion that emphasizes fault-tolerance and redundancy, given how much cheaper it will be to run those workloads on Graviton.
The chip arrives at a very interesting time in the server processor market.
Intel is at a crossroads, operating without a permanent CEO for the last five months and beset by manufacturing issues that could have an enormous impact on its ability to ship competitive products for the next year or so. AMD has an interesting new generation of server processors that AWS is offering to its customers, but has a long history of execution problems that have derailed promising products in the past.
In the past, betting on a chip maker was a multiyear investment given the all the work required to test and validate one’s software for new hardware platforms; once you had completed that painstaking task it was hard to justify doing it all over again for a new chip supplier without some healthy incentives.
But the 2015 decision to acquire Annapurna Labs might go down as one of the shrewdest investments in Amazon’s history, given the flexibility that team has given AWS. Not only did the Annapurna team design Graviton, it also paved the way for the chip to get up and running quickly within AWS’s cloud services thanks to its work on the Nitro chipset, a system of hardware and software that integrates a processor within a server.
Nitro’s abstractions give AWS the ability to switch new processors in and out of its hardware designs with far less qualification work than was once required, DeSantis said. That means it can react much more quickly to changes in customer needs than Intel or AMD ever could, given the granular data about usage patterns and world-class chip designers it enjoys in house.
“A lot of the general-purpose processors in the data center have acquired a bunch of features that were really meant to help enterprise data center operators run more efficiently (and) that aren’t necessarily the right solution in the cloud environment,” DeSantis said, noting how quickly things change on the cloud.
AWS isn’t ready to talk about how often it will refresh its new Arm line of processors, nor which contract manufacturer is building them for the company. But it seems quite likely that AWS will use the early days of A1 workloads to study how customers are using the chip, and even likelier that no one will be able to move faster to capitalize on new opportunities that emerge as more and more companies move into cloud computing and realize they don’t have to operate the way they always had in the past.
“We’re going to be able to move much more quickly to bring all sorts of hardware” to customers, DeSantis said. That has huge implications for this industry, and given that both Microsoft and Google also have in-house silicon talent at their disposal, a new arms race in the cloud might have just begun.
