Intel officials are expected to announce on Sept. 4 the availability of “Avoton,” the company’s next-generation Atom processor aimed at the growing ultra-low-power server space that rival chip maker ARM and its partners also have their eyes on.
The microserver space promises to be a heavily contested market as demand for such systems grow and Intel and ARM maneuver to address that demand. Intel officials have argued that the advantage is theirs: Avoton will be the second-generation follow-on to “Centerton,” it already has such data center features as 64-bit capabilities and greater memory and virtualization support, and offers x86-based tools and software that are familiar with most developers and data center administrators.
ARM executives have said their low-power system-on-a-chip (SoC) designs, which were initially created for the mobile device space, are a good fit for the microserver space, which is expected to continue growing, driven by such trends as cloud computing, big data and mobility. While the bulk of ARM-based server SoCs currently are 32-bit—from such vendors as Calxeda and Marvell Technology—ARM officials note that they expect chips using the upcoming 64-bit ARMv8 will begin hitting the market next year.
A company in the middle of all this is Applied Micro, a smaller vendor whose executives in 2011 announced they would come out with the industry’s first 64-bit ARM-based SoC by the end of last year. With the market heating up now and Intel and other ARM vendors outlining their plans for chips for dense, low-power servers, Applied Micro officials are looking to press what they say is an 18-month advantage over those competing ARM partners.
The company has built a 40-nanometer version of its X-Gene SoC, which executives call the first ARM server product on the market. During a conference call with analysts and journalists in July to talk about the company’s quarter financial numbers, President and CEO Paramesh Gopi said the company has completed testing a 28-nm X-Gene model in the company’s labs, and will tape out the part in late 2013 and begin sampling it to customers in the first half of 2014.
The 28-nm X-Gene also will be complemented by X-Weave, a family of connectivity products that will enable X-Gene to scale to 100 Gb/s and beyond. In addition, Applied Micro announced in April a partnership with Altera to develop joint solutions that will enable Applied Micro to leverage Altera’s field-programmable gate array (FPGA) technology in its products, including X-Gene.
The result will be an ARM-based server SoC that has the enterprise features and capabilities found in Intel’s Xeon server chips, Gopi said in an interview with eWEEK. It will be highly competitive with Intel’s x86 processors—a 30-year-old architecture built to support Microsoft’s Windows—and will be ahead of other ARM chip partners, who are essentially looking to build server chips using ARM’s Cortex-A57 designs, which Gopi said are found in tablets.
“We wanted to build a Xeon-class ARM” chip, he said, noting that other ARM partners are currently working on 32-bit parts. “There is no market for 32-bit in servers. It isn’t there.”
Not only is X-Gene a true 64-bit ARM SoC built exclusively for the server space, but it comes with features such as error-correcting code (ECC) memory, as well as I/O capabilities such as 40 Gigabit Ethernet, Generation-3 SATA and Generation-2 PCI-Express. Such features give Applied Micro a “sustained, differentiated advantage” over other ARM partners, Gopi said.
“If it’s not a Xeon-class chip, [enterprises] are not going to look at you,” he said.
Applied Micro CEO: X-Gene SoC Is X-Factor in Microservers
Web companies, such as Facebook and Google, which run massive data centers, are looking for small, low-power systems to run huge numbers of small workloads. The need in hyperscale environments are for systems that don’t take up much space or consume much power. Top-tier OEMs are looking to address the demand. Hewlett-Packard is beginning to roll out ultra-low-power systems from its Project Moonshot initiative, and while the first systems are powered by Intel’s Atom chips, future Moonshot systems will run on chips from ARM chip makers such as Advanced Micro Devices, Calxeda, Marvell and Applied Micro.
Dell, too, is offering microservers, including through its Copper strategy, which also will leverage ARM-based chips.
AMD executives, who will begin making ARM-based server chips next year, have said ARM could gain 20 percent of the server market by 2016.
Applied Micro’s stance illustrates the complex world of microserver SoCs. ARM and its partners are competing with Intel, aiming to gain traction in a server market that the giant chip maker has dominated for years. However, competition within the ARM community also will be fierce, as ARM partners look to grab a larger share than other ARM vendors. Gopi said Applied Micro’s head start gives it an advantage over other ARM partners. Meanwhile, AMD executives have argued that they expect their company’s expertise in making server chips and the strong relationship with OEMs will enable it to become the top ARM server chip maker.
Nathan Brookwood, principal analyst with Insight 64, said he is skeptical about whether Applied Micro will have an advantage over AMD and others. The problem is that company officials had promised a 40-nm chip with 64-bit capabilities and particular features by the end of 2012, Brookwood told eWEEK. However, some key features were left behind, such as an integrated fabric controller, which he said is a key capability for any microserver chip.
“If they had delivered on their schedule with the capabilities they talked about, they’d have a huge lead vis-à-vis other [ARM] vendors,” he said.
Having fabric controllers integrated onto the SoC makes for easier and faster communications between chips, server and racks of servers in the data center, Brookwood said. Both Calxeda, with its EnergyCore SoCs, and AMD with its Freedom Fabric, are integrating fabric controllers into the 64-bit ARM SoCs they are planning to launch in the middle of next year. Intel, with its evolving Rack Scale Architecture efforts, is looking to integrate fabric management onto the silicon, so any advantage Applied Micro might have had has been neutralized, he said.
“It’s going to be a ballgame,” Brookwood said. “Intel shouldn’t be counted out.”