AWS CEO Andy Jassy introduces Graviton2 during his keynote on Tuesday at re:Invent 2019. (AWS Photo)
If Amazon Web Services’ first Arm server processor was a shot across Intel’s bow, its second-generation Graviton processor is something much more threatening.
When it arrives in production in the first half of 2020, Graviton2 will become the most powerful processor option among AWS’s general-purpose compute instances, with up to a 40 percent performance increase for some workloads compared to the current-generation Intel and AMD processors currently available. It will also be significantly cheaper -- up to 20 percent -- to run workloads on Graviton2 compared to the traditional options, said Raj Pai, vice president of EC2 product management at AWS, in an interview at re:invent 2019.
The first Graviton processor was as much a proof-of-concept product as anything, a statement by AWS that the years of wondering whether or not Arm servers would ever gain a foothold in the data center were coming to a close. Intel controls around 95 percent of the server processor market, and server buyers like AWS have long itched for an alternative supplier, especially as Intel ran into problems over the past few years with manufacturing delays and its response to the security issues caused by the Meltdown/Spectre chip design flaws.
Arm doesn’t make chips; instead, it lays out designs for processor cores that other companies like AWS’s Annapurna Labs use to design and manufacture their own processors. If you own a smartphone, you’re using a chip designed in part by Arm, which has enjoyed the vast majority of the mobile phone market for more than a decade.
Earlier this year, Arm rolled out its Neoverse N1 core, its first processor core designed for “infrastructure-class” servers, said Mohamed Awad, vice president of Arm’s infrastructure business, in an interview at re:Invent. Graviton2 is based on that processor core, which depending on the configuration can deliver a 60 percent performance improvement compared to the core on which the original Graviton processor was based, he said.
An AWS data center (AWS Photo)
Work was already underway on Graviton2 when the first version was announced last year at re:Invent, Pai said. At the time, AWS assigned the first Graviton chip a special instance type -- A1 -- as to distinguish its performance from the more powerful and more widely used M, C, and R general-purpose compute instances that AWS offers on its flagship EC2 cloud compute service.
However, the second-generation Graviton chip takes a big step forward. The original chip had 5 billion transistors, while the second-generation chip has 30 billion transistors, and it was designed for the most advanced 7-nanometer chip manufacturing techniques currently available. Intel, which has been considered the best chip manufacturing operation on the planet for decades, has struggled to get its 7-nanometer manufacturing process into production.
Graviton2 also scales up to 64 virtual processing units, four times the number of processing cores on the original chip. That gave AWS the confidence to launch Graviton2 in the mainstream M, C, and R instance families, setting up a very intriguing choice for AWS customers deciding where to run new applications.
“It became pretty clear, pretty quick, that (Graviton2) would offer a real value proposition for customers across mainstream workloads, just because of the sheer amount of price performance (gain) we could get,” Pai said.
Intel will likely have new server processors out toward the end of 2020, and AMD continues to revive its data center ambitions, so the performance arms race will continue. But with Graviton2, AWS has established that Arm server processors are ready for mainstream enterprise computing users, and it has also shown that it can move quickly to introduce new designs.
“It is still early days, obviously, but if even half of the major hyperscalers and cloud builders follow suit and build custom (or barely custom) versions of the Arm Holdings Neoverse chip designs, which are very good indeed and on a pretty aggressive cadence and performance roadmap, then a representative portion of annual X86 server chip shipments could move from X86 to Arm in a very short time – call it two to three years,” wrote Timothy Prickett Morgan, co-editor of The Next Platform, in a deep dive report on Graviton2 and how it fits into the broader chip ecosystem published Tuesday.
Performance had always been one of the bigger hurdles behind the slow adoption of Arm chips in the data center, but software support for the chips (which are not compatible with Intel or AMD’s chips) also needed to improve before enterprises would consider building production applications around the technology. The original Graviton processor helped spur independent software vendors like Red Hat, Docker, and Ubuntu, among others, to support Arm chips, and Pai thinks the second-generation chip could really accelerate that support from the software community.
AWS customers are unlikely to port existing applications over to the Graviton2 processor, but given the cost benefits, it’s an interesting choice for new applications. And customers who are developing applications for tools like Kubernetes that span multiple computing environments might consider using the new instances as part of their deployment strategy, Pai said.
“We’re hearing a lot of excitement from our customers, and that to us is an indication that the timing is right for this sort of offering,” Pai said.
(Disclosure: AWS paid for my accommodations in Las Vegas.)