Talk of Intel and ARM Holdings usually centers on Intel’s efforts to make inroads into a booming mobile chip market currently dominated by ARM and its partners, including Qualcomm, Samsung Electronics and Nvidia.
However, Intel and ARM also are edging closer to competing in the small but growing market for microservers—small, highly efficient servers aimed at massive data centers that need a lot of systems to process huge numbers of smaller workloads while keeping the power consumption down. It’s an area of that ARM and its partners see as a good growth area for their low-power chip architecture that already is found in most smartphones and tablets.
But Intel has no intention of ceding that niche part of the server space to ARM, and is looking to its own low-power Atom chips to become the dominant platform in the microserver segment, an area of the industry that officials with the giant chip maker say they helped create several years ago.
In a recent conference call with journalists, Intel Fellow Matthew Adiletta talked about Intel’s six-plus-year journey in defining the microserver market and creating an Atom chip for microservers. Adiletta in 2006 was asked to meet with a CTO at a Wall Street firm that was using blade servers. The meeting tuned him into the need for greater density in data centers as a way to reduce the amount of cabling, improve infrastructure management and get compute resources online more quickly, he said.
Research and development into helping create such density began in 2007, including the development of integrated CPU cards based on Atom and Core2 Duo chips. For external validation, Adiletta took it to Andy Bechtolsheim, a co-founder of Sun Microsystems and currently a co-founder and chairman at Arista Networks. Bechtolsheim asked questions about everything from power to heat to performance, then put his head in his hands, according to Adiletta.
After a few moments of silence, Adiletta quoted Bechtolsheim as saying the following: “Geez, it just hurts my head to think about all of the opportunities this could provide if we can realize it.”
More work with researchers at Carnegie Mellon University followed in 2009, and Intel officials started talking about the kinds of workloads—Adiletta pointed to Hadoop as an example—that could work well within an Atom-powered microserver. Such workloads run many small but highly parallel “chunks” of code, he said. A server running a dual-core Atom chip could help process such workloads.
“Having more chefs in the kitchen helps … depending on what is being served,” Adiletta said.
Adiletta’s talk with journalists comes as Intel is preparing to launch the Atom S Series “Centerton,” the first of its Atom chips for servers that is scheduled for release this month. That will be followed next year by the next generation of the Atom S Series, “Avoton.”
Intel Pushes Microserver Vision as It Readies Release of Centerton
Adiletta said Intel sees a strong marketplace developing around these dense systems and that Intel intends to be a leader in the space.
“Intel has a head start in microservers, we are very bullish and have an excellent road map,” he told the journalists.
Centerton already is giving Intel an advantage over ARM, which isn’t expected to see chips with its upcoming 64-bit architecture hitting the market until 2014. In November 2011, Hewlett-Packard announced that it was partnering with Calxeda, which makes ARM-based server chips, to create ultra-low-power servers as part of its larger Project Moonshot initiative. However, HP officials in June announced that the first servers from Project Moonshot will be based on Centerton, with the initial systems shipping by the end of this year.
Intel executives have argued that Atom, based on the same x86 architecture as the company’s other processors, has a host of advantages over chips based on ARM’s architecture, including the ability to run many legacy software applications and enabling customers to use familiar tools for the Intel Architecture. In addition, it won’t be until ARM’s upcoming chips that key server capabilities—including 64-bit capabilities, better virtualization support and greater memory—are included.
However, when ARM does come around with its ARM64 architecture, there will be myriad chip makers to churn out the chips, including Calxeda, Marvell Technologies, Samsung—which looks to be putting together a strong server chip team in Texas—and even Advanced Micro Devices, which has competed against Intel for years in the x86 chip space.
AMD executives announced in October that along with its traditional x86-based Opteron processors, the company also will begin making 64-bit ARM-based server chips. AMD also is taking a strong interest in the microserver category, with officials calling such dense systems a key part of their turnaround plans for the company. In addition, AMD in February bought SeaMicro, which makes microservers.
Some top-tier server OEMs also are planning to offer ARM-based systems. Along with HP, Dell is partnering with Marvell to create their Copper servers, which will be powered by Marvell’s Armada XP chip.