SANTA CLARA, Calif.—Applied Micro’s next-generation ARM-based X-Gene processor will be a more robust server chip that will offer significant performance gains over its current offerings and will target data center workloads now running on Intel’s mid- and high-level silicon.
Applied Micro President and CEO Paramesh Gopi introduced the X-Gene 3 during the second day of ARM’s TechCon 2015 show here Nov. 11, saying that when the system-on-a-chip (SoC) starts shipping in systems in 2017, it will be the only ARM-based chip that can challenge Intel’s dominance in the data center.
“There will be no higher-performing ARM part in the market,” Gopi said in an interview with eWEEK.
X-Gene 3 will come with 32 cores running up to 3GHz, eight DDR4 memory channels running up to 2,667MHz and 42 PCIe Gen 3 lanes. The SoC, manufactured via Taiwan Semiconductor Manufacturing’s 16-nanometer FinFET process, will offer four to six times the performance of the current X-Gene chips, Gopi said.
Applied Micro’s unveiling of the X-Gene 3, combined with statements made by enterprise and tech vendor officials here about their plans for using ARM-based server chips, is putting some muscle behind ARM executives’ bold contention that a quarter of servers sold in 2020 will be powered by ARM-based chips. Intel currently owns about 95 percent of the global server market.
Patrick Moorhead, principal analyst with Moor Insights and Strategy, noted that there have been some OEMs that are using ARM-based processors in some systems and reports of proofs-of-concept (POCs) running inside some enterprises, but he had hoped to see more momentum behind ARM’s data center efforts by this time.
“We still haven’t seen mass deployments,” Moorhead told eWEEK. “But I feel more comfortable [after Gopi’s announcement and ARM customer statements] that we will see deployments, and they will happen soon.”
ARM-based SoCs from the likes of Qualcomm and Samsung can be found in the bulk of mobile devices—particularly smartphones and tablets—on the market today, but the company in the last few years has been pushing to extend the reach of its silicon into new areas. That has been on display during TechCon, with much of the focus on the Internet of things (IoT). However, there also has been a lot of talk about ARM’s efforts to gain a foothold in the data center, and the executives’ contention that they can capture 25 percent of the market—a statement they made to investors earlier this year—has gotten a lot of attention and met with some skepticism.
Some of that skepticism has come from the fact that ARM officials have been talking about this for several years—then-CEO Warren East said in 2010 that the company would begin cutting into Intel’s market share by 2014—with relatively little to show for it. Applied Micro has led the way in developer ARM-based server SoCs based on the 64-bit ARMv8-A architecture, though Advanced Micro Devices and Cavium also sell server chips (though Cavium has seen more success right now in networking), and others—such as Qualcomm and Broadcom—have plans in the works.
Some vendors also are adopting ARM chips for some of their products. Hewlett Packard Enterprise (HPE) leads the way with its X-Gene-based ProLiant Moonshot m300 low-power compute module, though Dell and others—such as Cirrascale, E4 Computer Engineering and startup SoftIron (which had a system powered by AMD’s ARM-based Opteron A1100 chips on display at the show)—also are pushing hardware. Supercomputer maker Cray also is testing ARM-based chips.
Regarding end users, Applied Micro announced earlier this year that PayPal has deployed ARM-based servers in its data center, though most other efforts fall into the PoC or test categories.
Applied Micro Unveils Next-Gen ARM Server SoC
At the show, ARM officials warned about focusing too much on what is known publicly, adding that there is a lot going on behind the scenes. In addition, opportunities in such areas as hyperscale computing, cloud environments, high-performance computing (HPC) and the China market are significant, the company continues to sell licenses of its designs and interest in ARM’s architecture is growing, they said.
“I’m comfortable with where we are at this point,” Lakshmi Mandyam, director of server systems and ecosystems at ARM, said during a roundtable discussion with journalists and analysts. “There are a lot of proofs-of-concept going on with ARM.”
Mandyam also noted that the open-source software ecosystem around ARM’s server designs continues to expand, both through the efforts of ARM as well as through industry groups like Linaro. The Ubuntu and CentOS Linux operating system distributions run on the ARM architecture, though Red Hat continues its support in a developer release.
During the panel discussion at TechCon, officials with both HPE and EMC said they were building storage systems that will run on ARM-based processors and that both are testing SoCs in their labs. Mike Robillard, an architect and distinguished engineer with EMC, said his company has used ARM products in the past. The bulk of its systems run on Intel chips.
“This sort of feels like the next step in this relationship,” Robillard said. “We fundamentally believe that choice is beneficial.”
Dave Preston, distinguished technologist at HPE, noted that in lab testing, using X-Gene with four solid-state drives (SSDs) provided 200,000 IOPS (input/output operations per second).
“That’s quite compelling for a low-power chip,” Preston said.
Bert Shen, vice president of technology business development at Morgan Stanley, said his company is testing an application on an X-Gene-based Moonshot system. The Wall Street business is looking to reduce costs as well as find a second supplier to Intel.
“There are a number of users in the same position as us,” Shen said.
ARM officials also noted that they are running their ARM.com Website on ARM servers, giving the company a proof point for the architecture, according to CEO Simon Segars.
“If we want other people to do this, let’s do this ourselves,” Segars said in a roundtable discussion with journalists. “ARM.com just works, and that’s how it should be.”
The new capabilities of ARM-based SoCs, combined with the direction of enterprise computing, are working in ARM’s favor, according to Nathan Brookwood, principal analyst with Insight 64.
“I’m reasonably optimistic about the progress,” Brookwood told eWEEK. “Obviously, it’s taken a lot longer than anyone expected when they started talking about 64 bits.”
That said, the rise of hyperscale and cloud-scale computing has put greater emphasis on such metrics as power consumption, energy efficiency and open-source software, which plays to ARM’s strengths, said both Brookwood and Matt Eastwood, senior vice president of IDC’s Enterprise Infrastructure and Datacenter Group. In addition, end users are always looking for a viable challenger to Intel to help drive innovation and cost savings and to give them a second source of silicon. IBM, through the OpenPower Foundation, also is pushing to have Power be that alternative.
Applied Micro Unveils Next-Gen ARM Server SoC
Moor Insights and Strategy’s Moorhead also said that while ARM keeps pushing forward, so does Intel. The vendor continues to drive down the power consumption of its processors, is expanding its custom chip capabilities, is growing its use of accelerators like Xeon Phi and field-programmable gate arrays (FGPAs) via its planned Altera acquisition, and earlier this year launched the low-power Xeon D, the company’s first server SoC.
“Intel keeps plugging product holes,” Moorhead said.
Intel also is driving development of its upcoming “Skylake”-based “Purley” midrange Xeon E5 and high-end E7 server chips, which reportedly will come with improved performance, a new memory architecture, an integrated network fabric and a range of integrated accelerator options.
Those are the chips that Applied Micro’s X-Gene 3 SoC will compete with, CEO Gopi said. When ARM first started talking about getting into the server space, the focus was on low-power microservers. But Gopi and officials with Cavium said they were targeting the mainstream server space with their processors. (Cavium offers ThunderX.)
The performance improvements in the X-Gene 3—which has been under development for two years—will be driven in large part by a new interconnect technology Applied Micro has developed, Gopi said. The interconnect will enable “many cores to scale linearly” and unite 256 cores with 2TB of memory each. The CEO said he will disclose more about the interconnect technology—which he said will outperform Intel’s QPI and not be proprietary—at the SC 15 supercomputing show in Austin, Texas, which starts Nov. 15.
The pressure on Applied Micro now is ensuring that X-Gene 3 stays on schedule and delivers what it promises, Moorhead said. The company will begin sampling the chip in the second half of 2016, with expected systems running on it coming on the market in 2017.
Gopi said the company waited until now to talk about X-Gene to ensure that it could pass internal testing. He also pointed out that it’s based on two other generations of what he called “battle-hardened” SoCs: X-Gene 1 SoCs are shipping now, and the company is sampling X-Gene 2. He noted that the vendor has shipped a $1 million worth of X-Gene products as of June, and that the SoCs were used in 10,000 servers that were shipped in a recent quarter.
The X-Gene 3 will be the key driver behind ARM’s push to 25 percent market share by 2020, Gopi said, adding that “this is the only way we’re going to get to those numbers.”
“We’re very, very optimistic about next year,” he said. “This is no longer a PowerPoint exercise. This is not an exercise where we have aspirations. This is reality.”