Advanced Computing in the Age of AI | Wednesday, July 24, 2024

Arm Unveils Cloud-to-Edge Internet Infrastructure 

Source: Arm

Arm today announced its Neoverse cloud-to-edge internet infrastructure foundation, designed to support a world in which a trillion intelligent devices will transmit data to and from the internet.

Unveiled at the Arm TechCon conference in San Jose, Arm offered a look at its Neoverse processor IP roadmap and infrastructure-class IP for 5G networks designed to enable more ubiquitous compute from the cloud to the edge. The roadmap begins with the launch of the “Ares” IP platform early next year on 7nm and delivering what Arm said will be gains of 30 percent per generation through 2021.

Neoverse is backed by an ecosystem of nearly 40 companies that includes HPE, Sprint, Qualcomm, Microsoft Azure, Ampere, Xilinx and Cisco, and it’s been endorsed by Kubernetes, SUSE, Python, Docker and Redhat, among others.

In a pre-announcement press briefing last week, Drew Henry, Arm’s SVP/GM, Infrastructure Line of Business, said Neoverse is focused on the internet infrastructure, “what’s deployed in corporations, in how you handle backhaul, what gets deployed in large data centers,” including top-of-rack switches, gateways, WAN routers, cellular base stations and servers, “what’s considered the core infrastructure used for powering the internet.”

Arm Neoverse Model

Backed by findings from market watcher IDC released last June, Arm said it is the largest architecture deployed in the global internet infrastructure with nearly 30 percent unit share. Henry emphasized the growing demands on the internet to more efficiently move larger volumes of data, particularly video data, which he said comprises 70 percent of data transmitted on the internet today.

“We’re uniquely positioned as a company (to build this infrastructure) because we’re involved in discussions with every major leading company in the world trying to decide how to deploy these types of systems,” Henry said, “how do you put in smart parking systems, smart retail systems, smart transportation systems, autonomous driving systems.”

“This diverse ecosystem and solution that we offer, it’s really built around figuring out how we give all the flexibility our partners need to design whatever it is they want to design,” said Henry. “We’ve got all of our SOC partners and they have a variety of design options, they can use our IP, they can use their own architecture designs, they can build in accelerators they want to add into these things, they can build around different implementations of memory, different implementations of I/O interfaces, they can use whatever boundary interfaces they want to use, they can build on top of a software compliant platform. We offer our partners the design flexibility options they need to bring these Neoverse-class products to market. I think this is key in this next number of years as we begin to think about what is the right type of silicon models, system models, software models necessary for this new, developing world of infrastructure.”

The Neoverse roadmap will proceed from the 16nm Cosmos platform of today to 7nm Ares next year, 7nm+ Zeus the year after and 5nm Poseidon in 2021.

“The modern datacenter is no longer a physical construct, but a center of data and compute residing in the cloud and on the edge,” said Patrick Moorhead, principal analyst, Moor Insights & Strategy. “More than ever, organizations must consider distributed, connected datacenter design methods to support the data and devices coming in the 5G world. Arm is one of those rapidly emerging in the market and with Arm Neoverse purpose-built IP, it should be well-positioned to support many of the compute spectrum needs from hyperscale to edge access.”

Gaurav Singh, VP, architecture and verification, Xilinx, added that “High-performance IP, along with a complete ecosystem, enables customers to take full advantage of the flexibility inherent in our Arm-based products. The evolution of these cores, coupled with the capability of CCIX (fabric), provide an ideal platform for smart offload and purpose-driven edge compute platforms.”