eWEEK 30: Intel is playing catch-up in the low-power mobile device market where ARM is its strongest rival. Now ARM is taking aim at Intel's bread-and-butter data center server market.
Editor's Note: This article concludes eWEEK's series of retrospective articles that look back at 30 years of personal computer history—from Feb. 28, 1984, the publication date of the first issue of PC Week, the print predecessor to eWEEK. This final feature brings us to the 30th anniversary of PC Week's debut as eWEEK continues to cover the latest computer industry news in the era of mobile devices, cloud computing and social networking. As we mark this anniversary, eWEEK's editors and writers hope you have enjoyed this look back at computer history, and we look forward to serving you, our readers, for at least the next 30 years.
Andrew Feldman has looked at the changing computing landscape around him and has come to a conclusion: a single processor architecture in the data center just won't work anymore.
As corporate vice president and general manager of Advanced Micro Devices' Server Business Unit, Feldman has an interest in both sides of the issue. AMD for decades has made the bulk of its money from selling processors based on the x86 architecture and designed for servers and PCs, competing against Intel, the world's largest chip maker.
At the same time, Feldman and other AMD officials have been among the most vocal proponents of ARM's ambitions to leverage the growing demand for energy efficiency in the data center and bring its low-power system-on-a-chip (SoC) designs now found in most smartphones and tablets into highly dense and very small microservers.
The rapid changes going on in the computing industry—from the massive numbers of mobile devices connecting to the Internet to the rise of cloud computing to the growth of hyperscale data centers—present an opportunity for a new chip architecture to challenge Intel and its x86 dominance in servers, he told eWEEK last year.
"This new environment is going to have new needs, and the same-old, same-old will not work anymore," Feldman said.
It will take a while to see whether Feldman is right. Systems powered by 64-bit ARM-based processors aren't expected to start hitting the market until later this year and into 2015. Meanwhile, ARM and its partners face a range of challenges as they try to carve off a bit of Intel's vast market share in the server processor space.
That said, 2014 is expected to be a busy year in the growing competition between Intel and ARM, a contest that is playing out on multiple fronts—most notably mobile devices, the data center and the burgeoning Internet of things (IoT).
"Basically, both Intel and ARM are pushing toward a goal of [ensuring that] products are based on a single architecture," Charles King, principal analyst at Pund-IT Research, told eWEEK. However, how they go about working toward that goal is a key difference between the two.
Intel and ARM have been circling each other for several years, each one sitting in a position of strength in one segment of the tech industry while jealously eyeing the other's dominant position in their respective markets. Both companies covet the growth opportunities that are possible if they can only take away some of that business from their rivals.
Intel is far and away the world's largest chip maker, aggressively keeping pace with Moore's Law and building its innovation and manufacturing capabilities over the past two decades to hold market shares that range from more than 80 percent to more than 90 percent. Intel, an enormous and well-funded company, has thousands of engineers that design the chips, which are then manufactured in Intel's state-of-the-art fabrication facilities.
The company also has robust and growing software and development units that offer programmers, OEMs and end users a full software ecosystem that is based on familiar Intel technology that is found in the bulk of the PCs and servers they use.
"Intel argues that IT environments have become so increasingly complex that having a single vendor in charge of semiconductor development and manufacturing is the best way to … experience a fully optimized experience for their customers and end users," King said.
What the company needs is a stronger position in the booming mobile device market, where more than 95 percent of all smartphones and tablets—from Apple's iPhone and iPad to almost every device running Google's Android operating system—run on SoCs designed by ARM.
Those designs are licensed to the likes of Samsung and Qualcomm, who put their own IP on top of the architecture and sell the chip. ARM's Connected Community program boasts more than 1,000 partners in everything from design to manufacturing.
It's a model that ARM and its proponents argue promotes competition and speeds up innovation. ARM officials said the 50 billionth ARM-based chip shipped in the 2013 fourth quarter. Now ARM wants to move its technology up the ladder and into the PC and server spaces to chip away at Intel's massive market share.
"It's a very different approach," Laurence Bryant, vice president of strategic marketing at ARM, told eWEEK. Intel's model "is very vertically integrated … where we're very horizontal. [ARM's approach is] about delivering the best experience, the lowest power and the highest integration with partnerships."
Through the vast array of partners, ARM can achieve a level of integration—with other companies putting their own technologies atop ARM's—that is difficult to do as a single vendor, Bryant said.
Working with partners, including tech heavyweights like Samsung and Qualcomm, is also a way for ARM—which last year for the first time generated more than $1 billion in revenues—to spread the workload and to give it a chance to compete with Intel, which in 2013 saw $52.7 billion in revenues.
"ARM is an extremely innovative and accomplished company, and it's proving effective through a series of collaborative partnerships," King said.
So each company is looking to move from a position of strength into new markets heavily controlled by the other one, while at the same time eagerly eyeing the nascent, relatively wide-open and potentially very lucrative IoT space. And both sides see 2014 as a significant year in their efforts.
Executives at Intel—like many of their brethren at other established tech vendors like Hewlett-Packard, Dell and Microsoft—were slow to recognize the impact smartphones and tablets would have in the industry as a whole and the PC market in particular.
While serving as CEO, Paul Otellini said several times that although tablets were a fast-growing part of the PC market, they essentially were "additive" devices, complementing the more robust traditional notebooks users relied on. However, as the global PC market contracted over the past couple of years—sales fell off 10 percent in 2013, according to Gartner—Intel has ratcheted up the innovation around its low-power Atom platform, enhancing performance while driving down power consumption in an effort to offer OEMs an alternative to ARM.
Atom chips also can support Android as well as Microsoft's Windows, an important step given that Android is found in the bulk of mobile devices.
A significant step came last year when Intel unveiled its 22-nanometer Silvermont SoC micro-architecture that company executives said meets or exceeds ARM's architecture in performance and energy efficiency. Earlier Atom chips had failed to gain Intel much traction in mobile devices.
"We're breaking the myth that ARM can do things that Intel cannot," Dadi Perlmutter, then executive vice president, general manager of the Intel Architecture Group and Intel's chief product officer, said last year at Silvermont's launch.
The Atom Z3000 Bay Trail SoCs launched in September 2013 sport the Silvermont architecture and are aimed at tablets and new form factors, such as two-in-one systems that can be used as a tablet or notebook. At the Mobile World Congress 2014 show on Feb. 24, Intel unveiled the 64-bit Atom Z3480 Merrifield chips for smartphones. Silvermont and Merrifield are 22nm chip families, and Intel is already working on the next 14nm version.
Also at the show, Intel officials said the company was partnering with Lenovo, Asus, Dell and Foxconn to put the company's chips in smartphones and tablets in 2014. CEO Brian Krzanich also has said that he expects hardware makers to deliver 40 million tablets equipped with Intel chips by the end of the year.
Whether that comes to pass remains to be seen, but Pund-IT's King said Intel has strengths it can rely on. Often in mobile, the debate is technological, with the critical point for chips in the space to be able to support highly computational processes while also being very energy-efficient. However, there's also a financial side to it.
"Yes, device makers do want chips with these features, but if Intel can create a chip that offers 90 percent of ARM chip capabilities and sell them at a significant discount," that may be good enough for some OEMs, he said.
Intel also is being smart in targeting emerging markets for devices running on their chips rather than aiming for regions like the United States and Western Europe, where smartphones and tablets are closer to a saturation point, he said. King also noted that Intel has shown it can play "the long game." He noted that when AMD released the 64-bit x86 Opteron server chips in 2003, it captured about a quarter of the server market before Intel could sufficiently respond almost two years later. Now Intel's server market share is north of 90 percent.
However, that was a server market that Intel already dominated at the time. Intel's challenge in mobile is different because it is the underdog fighting a dominant market leader, an unfamiliar position for the chip giant. Still, Intel comes with resources and talent and its current market position can be helped by the typical 18- to 24-month turnaround time for new smartphones and tablets. The churn rate will keep the demand high for the devices, giving Intel a better opportunity to get a solid footing in the market, King said.
The potential rewards are high. If Intel could establish itself as a strong second in a market where almost 1 billion smartphones were shipped last year, it could translate into billions of dollars in sales, he said.
For its part, ARM is trying to muscle into Intel's territory in both PCs and servers. ARM has seen some success in the PC market; Hewlett-Packard, Samsung, Dell and other OEMs are rolling out Chromebooks that are powered by ARM chips. However, Microsoft's ARM-based Surface RT Windows tablets have had trouble in the marketplace, with consumers preferring the Surface devices that run on Intel chips.
Now ARM is gearing up to launch its ARMv8-A architecture in 2014, which will include data center features like 64-bit computing, more memory capacity and greater virtualization support. Officials with ARM and their supporters have talked for several years about the opportunity available in the data center, particularly given the growth of hyperscale environments run by the likes of Google, Facebook and Microsoft, who run thousands to millions of servers that process massive numbers of small tasks. In such dense data centers, energy efficiency is often more important than sheer performance.
This is where ARM officials believe they can gain some traction. They have some top-tier OEMs in Dell and HP that are looking to put ARM-based chips into new microservers they are developing. An array of chip makers—including AMD, Samsung, Qualcomm, Nvidia, Applied Micro and Marvell Technologies—are working to leverage the new architecture for server SoCs.
They also have a growing number of software makers—from Microsoft and Red Hat to SUSE and Canonical—throwing their support behind ARM by helping create a software ecosystem to support the servers when they begin launching later this year, according to Lakshmi Mandyam, director of ARM's Server and Ecosystems unit.
The effort also is being supported by the Linaro open-source project, which helps develop software for ARM chips and has taken an important role in helping optimize Linux for the architecture.
At HP's launch of its first Moonshot microservers last year—which currently run on Intel processors—Mandyam said development of the ecosystem will be crucial to ARM's efforts and that the growth of open source in data centers played to the company's strengths.
"Open source is the great equalizer," Mandyam said. "I don't think the gap [between ARM and Intel in server processor technology] is as much as you might think."
As 2014 unfolds and rolls into 2015, ARM expects momentum to build, she told eWEEK. Already this year, there have been significant steps. AMD in January unveiled its eight-core Opteron A1100 ARM-based SoC, which officials said will begin sampling this quarter. In addition, HP officials said they will begin shipping ARM-powered Moonshot servers this year. Meanwhile Dell earlier this month unveiled an ARM-based proof-of-concept system aimed at helping expand the ecosystem around the SoC architecture.
Just as significant, ARM and some partners in late January rolled out the ARM Server Base System Architecture (SBSA) specification, a platform standard for servers running 64-bit ARM chips.
Dell has been working with ARM since 2008 to bring the SoC architecture to the data center, according to Robert Hormuth, senior distinguished engineer and executive director of platform technology and architecture at Dell. Hormuth expects things to speed up throughout 2014.
"Now that 64 bit [in the ARM architecture] is real … expect to see some development platforms come out, probably at a [rapid] pace," he told eWEEK. "All the pieces are starting to come into play."
However, there will be challenges, according to Pund-IT's King, with Intel being the largest. While ARM and its partners are still several months away from seeing the first ARM-based servers get to market and a year or more away from broad deployment by end users, Intel is on the second generation of Atom SoCs for low-power servers, and is already powering commercial HP Moonshot systems. The company also is working on the 14nm "Denverton" Atom chip, due out later this year.
"It's not a company that rests on its laurels very long, and it's a company that invests pretty massively in next-generation technology and manufacturing to keep that competitive edge," King said.
There's also the question of how big the server opportunity is for ARM. AMD's Feldman has said ARM could capture as much as 25 percent of the server market by 2019. However, others say the microserver segment could be 10 percent or less of the server space. That also will lead to a consolidation of ARM server SoC vendors, with King and others saying that eventually there will be room for three or four. Already one of the higher-profile vendors, Calxeda, shut down in December 2013 because it couldn't generate enough cash flow to remain in business.
"You're going to see a handful of players in the end, but there will be enough business to sustain that," Suresh Gopalakrishnan, corporate vice president and general manager of AMD's Server business unit, told eWEEK.
The technology is solid; the software environment around the architecture is growing strong; and the business demand is there, Gopalakrishnan said.
Whether ARM will be able to expand its reach beyond the Web-based companies and into enterprise data centers is another question, King said. The ARM architecture may be attractive to the likes of Google and Yahoo, but it will be difficult to sell ARM-based servers to enterprises if ARM and its partners are looking in that direction.
Most large enterprises tend to run systems that are powered by bigger CPUs and are highly virtualized, with the chip resources used to host virtual machines, he said. ARM-based systems more likely run as clusters, with small systems connected to robust backplanes.
In addition, enterprises will probably be less inclined to bring a new architecture and all the attendant training and recompiling that comes with it into their data centers, King said. He also said that Intel's single architecture approach fits better with enterprises. He cited the case of Unix systems in the 1980s, when IBM, HP and Sun Microsystems all offered their own variants of the RISC chip architecture, with companies having to decide which one they would invest billions of dollars in.
ARM could find itself in the same situation, with three or four vendors putting their own customized ARM-based chips on the market. For enterprises, the simpler and safer way to go would be with Intel, with a single architecture that it continues to invest in.
While both Intel and ARM are looking to push their way into markets that are dominated by the other, one area that is opening up for both is the Internet of things, where billions of intelligent devices connect to the Internet and communicate with people and each other. Gartner analysts said that by 2020, the IoT will include as many as 26 billion devices and have $300 billion in sales. Cisco Systems officials say the IoT could be worth $14.4 trillion in profits to businesses worldwide by 2020.
"The Internet of things is wide open right now," King said. "We're really in the early days there. From an endpoint perspective, ARM's energy efficiency makes it a natural leading player in that part of the market, but Intel's got some assets that will serve it well in this space."
ARM officials in March 2013 announced the company's energy-efficient Cortex-MO+ chip for the IoT, and five months later, ARM bought Sensinode Oy, which makes IoT software. CEO Simon Segars said last year that the "Internet of things runs on ARM. … Our technologies provide the functional building blocks in a huge range of products, including cars, heart monitoring systems, washing machines and lighting. Energy efficiency and miniaturization are essential in these technologies."
For its part, Intel in September 2013 introduced the Quark family of chips, which are a fifth the size of Atom SoCs and consume a tenth the power and are aimed at the IoT. In November of 2013, Intel created an Internet of things business unit, which not only will include Quark and Atom hardware, but also software from subsidiaries Wind River and Intel Security (formerly McAfee).