British chip designer Arm Ltd. today announced a more advanced mesh interconnect for the Neoverse architecture chips that it said is a key element for partners aiming to develop more sophisticated systems-on-chip.

Arm announced its Neoverse V1 and Neoverse N2 platforms in September, saying the chip blueprints will serve as the basis for a new breed of more powerful, server-grade central processing units. At the time of their release, Arm said the V1 and N2 platforms were an upgrade on its existing Neoverse N1 architecture used by Amazon Web Services Inc. and three others that are counted among the world’s seven largest so-called hyperscale data center operators.

It’s important to note that Arm doesn’t build chips itself. Rather, it designs and licenses chip architectures to processor manufacturers, who use them as the basis of their own branded chips.

In a blog post today, Chris Bergey, senior vice president and general manager of Arm’s Infrastructure Line of Business, said the V1 and the N2 are compatible with the latest five-nanometer manufacturing process being adopted in the chip industry. They also support a number of other technologies that should give a boost to certain niche use cases, he said.

For example, the Neoverse V1 supports Scalable Vector Extension technology that delivers a higher per-core performance and better code longevity and provides SoC designers with more implementation flexibility. The Neoverse N2, meanwhile, supports SVE2, a newer version of the SVE technology that Bergey said drives more performance efficiency in cloud-to-edge use cases such as machine learning, digital signal processing, multimedia and 5G.

Bergey explained the Neoverse V1 and N2 chips offer up to 50% and 40% better performance, respectively, than the current-generation Neoverse N1. The V1 is designed for building processors that provide a high amount of performance per thread while the N2 is for CPUs with high core counts, he explained.

Which of these two chip types is preferable mainly depends on what applications a company is running: Software products that are billed per processor core, for example, are best run on processors with modest core counts and higher per-thread performance.

“As Moore’s Law comes to an end, solution providers are seeking specialized processing,” Bergey wrote. “Enabling specialized processing has been a focal point since the inception of our Neoverse line of platforms, and we expect these latest additions to accelerate this trend.”

Wikibon analyst David Floyer told SiliconANGLE that Arm’s Neoverse server chips are striking at the high-end of a data center server market, including high-performance computing, artificial intelligence and matrix application workloads that are currently dominated by x86-based chips sold by companies such as Intel Corp. and Advanced Micro Devices Inc. But he said that while more than 90% of servers in use currently run on x86 chip architectures today, those companies do have good reasons to be worried by the emergence of the Neoverse chips.

“I believe that Arm-based architectures will offer an increasingly lower cost, higher performance and greater security for cloud vendors,” Floyer said.

Analyst Patrick Moorhead of Moor Insights & Strategy told SiliconANGLE the Neoverse N2 is the most interesting of the two new designs thanks to its focus on general purpose server computing.

“I believe it will raise the bar on single-thread performance and will challenge both AMD and Intel on that vector,” he said. “V2 is interesting too as it’s focused primarily on the high-performance computing market where Arm has been very effective.”

Bergey said Arm’s partners have made strong progress as they work to implement the new Neoverse designs into their products. For example, Marvell Technology Group Ltd. recently announced a new family of Octeon networking chips based on Neoverse N2 and said it will provide its first samples by the end of the year. Marvell claims the Octeon chips will deliver a threefold performance boost over its previous-generation Octeon processors.

“Arm’s Neoverse N2 offers the industry’s top computing power per watt, delivering three times the compute performance and four times the SPECint per watt of the current generation,” said Raj Singh, executive vice president of Marvell’s Processors Business Group. “Our next-generation Octeon DPU family will leverage Arm’s Neoverse N2 cores to power critical infrastructure applications from 5G to storage to signal processing and security.”

The Neoverse chips are seeing broad adoption in the cloud too. For example, the Neoverse-powered AWS Graviton2 processors are rapidly expanding their footprint in Amazon Elastic Compute Cloud, while Oracle Corp. has said it plans to use Ampere Altra CPUs, based on Neoverse designs, in its Oracle Cloud Infrastructure, Bergey said. And Alibaba Cloud, the cloud computing arm of China’s e-commerce giant Alibaba Group, has shown some impressive results in benchmark tests of its upcoming Alibaba Cloud ECS Arm instances, showing 50% performance improvement versus older instances based on the Neoverse N1.

Another big Chinese tech firm that’s heavily invested in the new Neoverse architectures is Tencent Corp. Bergey said Tencent is currently working on both hardware tests and software enablement with a view to using Neoverse to power its cloud applications.

“These partners are taking full advantage of what is under the hood of Neoverse platforms,” Bergey said. “This is just the tip of the iceberg for both infrastructure workload benefits and on how our partners plan to implement and take Neoverse IP to market.”

Floyer told SiliconANGLE the major friction against moving data center servers to Arm’s architecture is the cost of migration. But he said cloud vendors own their data center platforms, and so they can bear the cost of migration and will most likely make the move first. “On-premises servers, except for cloud provider servers from AWS and others, will lag behind because individual ISVs will adopt Arm later than cloud vendors,” he said.

All in all, it would seem Neoverse has a bright future — so bright, in fact, that Arm is also pushing for the Neoverse designs to serve as the basis for a new breed of SoCs. To make that happen, it announced a new mesh interconnect that can be used to link the multiple components they house.

SoCs integrate multiple components such as the CPU, memory, input/output ports and secondary storage onto a single chip. SoCs are often seen in mobile devices and help to improve performance and reduce power consumption in such devices, though the designs do cost more to implement than single components.

Arm said the mesh interconnect is one of the most critical components of any SoC design. The new CMN-700 mesh interconnect is said to provide significant improvements on the previous-generation CMN-600, and enables what Bergey said is a “step function increase in performance on every vector,” from core counts and cache sizes to the number of types of memory and I/O devices that can be attached to the SoC.

“Arm has done very well in the recent years and continues to do so with the Neoverse N2 and V1 upgrades,” said Constellation Research Inc. analyst Holger Mueller. “Now all eyes are on adoption, and the news from Arm today in that respect is very encouraging.”

Photo: Arm