Getty Images/iStockphoto

Arm Neoverse roadmap targets enterprise infrastructure, cloud

Arm's roadmap for Neoverse V2 core is designed to handle 5G, HPC and edge workloads. Nvidia will incorporate the offering in its Grace data center chip in 2023.

Anticipating the needs of users and developers building cloud infrastructure that can handle high-performance computing, 5G and cloud-to-edge workloads, Arm Holdings unveiled the roadmap for its next-generation processor core.

Neoverse V2, code-named Demeter, contains the most recent V-series core along with the Arm CMN-700 mesh interconnect and several security enhancements. Its N-Series CPU is also in development and will be available to partners sometime in 2023.

"Compute workloads are on a relentless march higher, and becoming more complex," said Chris Bergey, senior vice president and general manager of Arm's infrastructure line of business, at a press briefing. "Machine learning and AI are taking over the future, and so infrastructure will look nothing like the past."

Over the next year, Arm will work closely with its cloud and software partners to optimize cloud-native software infrastructure, frameworks and workloads. These partnerships include contributions to projects including Kubernetes and Istio, along with several CI/CD tools used for creating cloud-native software for the Arm architecture.

Arm will also work to improve machine learning frameworks such as TensorFlow and a number of workloads such as big data, analytics and media processing.

The company is moving into more traditional enterprise spaces now, Bergey said, noting the work it has done with VMware on its Project Monterey and providing support for Red Hat's OpenShift and SAP's HANA.

"These cloud providers all use GPUs to underpin their cloud workloads, and the majority of them are using Arm," Bergey said.

Arm has built momentum in the enterprise the past two years, gaining the support of AWS, Microsoft and Google. What attracted the big three to the Arm architecture was the ability to build lower-cost, energy-efficient infrastructure for their cloud environments, although the tradeoff was lower performance than Intel and AMD chips could provide.

Users can go to AWS and run a workload on Arm-powered cloud servers for less than it would cost on Intel servers. But [Arm's chips] are only for pushing out web pages, not the heavier workloads Intel and AMD run.
Jack GoldPresident, principal analyst J. Gold Associates LLC

"Users can go to AWS now and run a workload on an Arm-powered cloud server for less than it would cost on Intel server," said Jack Gold, president and principal analyst at J. Gold Associates LLC. "But it's only for pushing out web pages and not the heavier workloads that Intel and AMD runs."

But with Neoverse V2, Arm hopes to compete more directly against Intel and AMD for cloud workloads where users put as much emphasis on higher performance as they do on energy savings.

Nvidia will use the power-efficient Neoverse V2 chip as the compute foundation for its Grace data center CPU. That, combined with the efficiency of LPDDR5X memory, will provide users with twice the performance per watt compared with servers powered by traditional architectures, according to Arm.

How Neoverse V2 performs relative to the x86-based offerings of Intel and AMD won't be fully known until Arm deliver the first silicon. In yesterday's briefing, the company didn't indicate when that would be, saying only that it hopes to deliver the product sometime in 2023.

Another hurdle Arm may face in going after Intel and AMD users is the lack of compatibility with their existing mission-critical applications, Gold said.

Jack Gold, president and principal analyst, J. Gold AssociatesJack Gold

"If you want to replace applications that were architected for x86 boxes, you have to make sure their applications are compatible with x86 code," Gold said. "And that is something that is not so easy to do."

Another analyst agrees the compatibility with existing x86 software is a problem that will take some time to fully address. But with the growing interest in running cloud-native applications, it's a problem that can be resolved.

"Hyperscalers have identified opportunities to build cloud-specific silicon that has now taken hold," said Dan Newman, a principal analyst at Futurum Research and CEO of Broadsuite Media Group. "Cloud isn't going to get smaller in the data center, and infrastructure isn't going to become less important."

Newman points to Apple's introduction of its M1 chip as one leading indicator of how a chip with some of the same compatibility issues with existing software overcame those issues.

"Apple raised concerns over compatibility issues with its chip, but over time optimized it [and] did a good job getting more applications to work with it," Newman said. "It won't happen as quickly for Arm with V2 because Apple has more control over its ecosystem than Arm."

In addition to the V2 and N2 unveilings, the company previewed the Neoverse E2 platform but offered few technical details.

Like the N2, the smaller-scale E2 will be compatible with the V2. With the E2, Arm has combined the Cortex-A510 CPU with the CMN-700 mesh, which gives server vendors more flexibility by offering an alternative CPU core to the N2.

As an Editor At Large with TechTarget Editorial, Ed Scannell is responsible for writing and reporting breaking news, news analysis and features focused on technology issues and trends affecting corporate IT professionals.

Dig Deeper on Data center hardware and strategy

Cloud Computing
and ESG