ARM Publishes 64-Bit Server Standard For Data Centre Push

ARM and its partners have revealed a platform standard for servers running 64-bit chips based on the company’s designs.

This is ARM’s latest move as it seeks to continue to challenge Intel in the data centre.

SBSA Spec

ARM officials on 29 January announced the ARM Server Base System Architecture (SBSA) specification, which gives system makers a framework for building systems powered by systems-on-a-chip (SoCs) built on ARM’s upcoming 64-bit ARMv8-A architecture. It also will fuel development of software for the systems, and help enable the ability to port workloads between ARM-based servers, according to ARM officials.

The announcement comes a day after Advanced Micro Devices, at the Open Project Summit in San Jose, California, unveiled its first ARM-based 64-bit chip, the 28-nanometer A1100 “Seattle” SoC. AMD, which has been among the most vocal proponents of ARM’s push into the enterprise, will begin sampling the four- to eight-core A1100 later this quarter, with expectations that systems running on the chip will begin appear later this year.

AMD was among the chip-making partners that helped ARM – which creates the chip architecture and then licenses the design to others, who put their own IP on top of it – develop the platform standard. Other participants included software companies like Microsoft, Red Hat, Citrix Systems, Canonical and Suse, and system makers Hewlett-Packard and Dell.

Platform Standard

It was important for ARM to come out with a platform standard that others can work off of, and also important to have the help of others in developing the platform, according to Jeff Underhill, director of server programs at ARM. Creating an active and expanding ecosystem is a key part of ARM’s strategy in challenging Intel’s dominance in server chips, and the development of the SBSA was an example of that involvement.

“The ecosystem understood that this was something we had to do, so they were motivated,” Underhill told eWEEK. “ARM’s business … is all about collaboration. If we’re to achieve [growing] ARMv8 servers in the data centre, we know that we need to work together.”

The SBSA specification includes the underlying SoC platform, firmware and network boot capabilities, the operating system and the hypervisor kernel, according to ARM officials. It standardises such low-level CPU and SoC attributes as timers, interrupt controllers and performance counters, and outlines minimum hardware requirements for firmware and OS vendors. It also stipulates that technologies should comply with industry standards for boot devices and that hardware be describable or discoverable, according to Underhill.

In a post on the ARM blog, he said that “silicon vendors are permitted to support capabilities beyond a given level as long as software created for that level is able to run unmodified. OS vendors are able to develop support for multiple levels in a single OS offering, thereby accelerating time-to-market and reducing maintenance by ensuring they can run across all ARMv8-A architecture-based server platforms.”

Partner Approach

ARM and its partners see an opportunity to gain traction in low-power servers with the growth of hyperscale data centre environments run by cloud providers like Amazon Web Services and Microsoft, and Web-based companies like Google and Facebook. In these facilities – which house huge numbers of servers the process massive amounts of data – power- and cost-efficiency are often more important than raw performance. ARM officials for the past few years have argued that the company’s history creating low-power SoCs for mobile and embedded systems fit that demand.

While some chip partners like Marvell Technology and Texas Instruments have offered some 32-bit ARM chips for servers, the industry won’t get a real gauge on demand until systems running 64-bit SoCs start hitting the market. Already Applied Micro has developed its 64-bit X-Gene ARM-based chip, and now AMD is getting ready to sample its first one.

However, ARM ambitions took a hit last month when Calxeda, an early pioneer in developing ARM-based server chips, closed its doors suddenly after running out of money, despite having raised millions of dollars and producing several 32-bit products.

New Workloads

Still, ARM proponents are confident. Speaking to reporters 28 January, Andrew Feldman, corporate vice president and general manager of AMD’s Server Business Unit, said new computing demands are creating a need for an alternative to x86 chips from Intel and AMD, and that he expects by 2019, ARM-based SoCs will account for 25 percent of the server chip market.

“These new workloads mean that the same-old way of doing things just doesn’t work,” Feldman said.

HP and Dell both have projects underway for offering low-power microservers powered by ARM-based SoCs.

However, Intel is also being aggressive in the low-power server space. The company in September 2013 launched its low-power 22nm Atom C2000 “Avoton” SoCs based on the new “Silvermont” microarchitecture that company officials say meets or exceeds ARM’s SoCs in performance and power efficiency. In addition, Intel later this year will release the next generation 14nm “Denverton” SoC.

In addition, while HP is promising ARM-based Moonshot server modules, the first ones that have rolled out are powered by Atom SoCs.

Originally published on eWeek.