Arm officials are rolling out the latest generations of the company’s Neoverse infrastructure platform that are designed to help enterprises get better control over the rapid changes happening in their data center with the rise of the internet of things, edge computing, 5G and other trends.
The company in October 2018 introduced Neoverse, an infrastructure-class platform that targeted everything from hyperscale data centers to edge computing environments that pulls together Arm’s own IP with products and technologies from third parties. Neoverse is aimed at a rapidly changing computing world that will soon include a trillion smart, connected devices, multiple clouds and massive amounts of data that will need to be collected, stored, processed and analyzed.
The data consumption model is changing from one where content is distributed via the internet—think Netflix and YouTube—to one that deals with information that is created at the network edge, where the IoT devices are located.
The platform unveiled last year, code-named “Cosmos,” was a 14- and 16-nanometer offering. Now the company is coming out with two new platforms under the “Ares” code name. The first is the Neoverse N1 platform, which comes with a range of infrastructure-class features, such as server-level virtualization, enterprise RAS support, power and performance management capabilities, and system-level profiling. It’s primed for performance and efficiency.
The other platform, the E1, is optimized for throughput, a key capability as 5G becomes more mainstream in the coming years. Both platforms are built on a 7nm manufacturing process and come with a range of reference designs and support for a broad array of third-party open software offerings.
“The point is, we’re not just providing a lot of IP,” Mohammed Awad, vice president of marketing for Arm’s infrastructure line-of-business, said during a recent two-day event in San Jose, Calif. “The point is, we’re providing an entire platform.”
Arm, long known for the small, power-efficient system-on-a-chip (SoC) designs that populate many of the smartphones and other small connected devices in the world, for most of the decade has been pushing to get its architecture into the data center and challenge Intel and its dominant position in the server chip market.
There have been stops and starts along the way—pioneer Calxeda went bust and Qualcomm pulled out after introducing its first Arm-based server chip—and there had been frustration by some industry observers at what they saw as the slow pace of development, with the frustration fueled at times by the inflated promises of Arm officials. However, the push has gained momentum in recent years. Cavium—now part of Marvell—has continued developing its Arm-based ThunderX line of chips, and others, like startup Ampere, also are creating data center SoCs.
OEMs like Hewlett Packard Enterprise have embraced Arm SoCs for some servers, cloud providers like Microsoft Azure are using Arm-powered systems in their environments, and VMware last year introduced a bare-metal ESXi hypervisor for 64-bit Arm servers. Huawei and Fujitsu are building their own Arm server chips, and Arm got a significant boost in November when top cloud provider Amazon Web Services (AWS) announced it is putting its own custom Arm SoC—dubbed “Graviton”—into new cloud instances.
‘Watershed Moment’ for Arm
Awad called the Graviton announcement “a watershed moment for us. … It was an incredible validation of all the innovation Arm has put in place.”
All of that coupled with the recent struggles by Intel—including delays in its 10nm Xeon chips—and the fast-changing data center environment has given Arm an opportunity to get its architecture into servers and other systems. Awad also noted the decline of Moore’s Law.
“Moore’s Law is coming to an end,” he said. “It’s slowing down. That means you can’t go with a single architecture [in the data center]. That’s yesterday’s model.”
With the N1 platform, Arm is introducing a compact and highly scalable architecture—it grew from four to 128 cores—which enables technology partners to add their own custom silicon to the chips to include such features as accelerators. The platform also includes a coherent mesh interconnect that also helps enable the compact design and diversity of on-chip capabilities.
Neoverse N1 offers a 60 percent improvement over the current Cosmo platform, with as much as 2.5 times the performance for some cloud workloads like NGINX. The E1 platform offers 2.7 times the throughput, 2.4 times the throughput efficiency and more than twice the compute performance of previous Arm generations. It also supports everything from a sub-35W base station to a multi-100GB router.
“While many people were initially skeptical about Arm’s claims for the server world, it’s clear with these announcements that they’re moving forward in a positive and aggressive way,” Bob O’Donnell, principal analyst with Technalysis Research, told eWEEK. “Not only is the range of different offerings impressive, the company’s focus on a number of often overlooked applications starts to make Arm a more serious player in the data center.”
O’Donnell also said that Arm is well-suited to address the growing edge computing environment, adding that the company’s multiyear roadmap is important as it pushes into the data center.
“Data center plans often occur over multiple years, so if Arm wanted to prove it was serious in this space, it needed to outline a long-term roadmap that shows both cloud providers and others interested in using Arm-based servers that they were in it for the long haul and have a logical progression of technology advancements moving forward,” he said.
Arm will release the next-generation platform next year with “Zeus,” which will be built on 7- and 5nm+ processes. “Poseidon,” a 5nm platform, will come in 2021. Arm officials noted that the plan is to have at least a 30 percent performance increase from one platform to the other, but added that Ares already is offering a 60 percent boost over Cosmos.