Intel’s Bryant: Chip Maker Expanding Reach in Data Center
Intel's Bryant: Chip Maker Expanding Reach in Data Center
Diane Bryant took over Intel's data center business at an interesting time in the industry.
The rise of cloud computing, virtualization, big data, mobility and other trends has rapidly changed the dynamics in the data center and expanded the field of competitors in a part of the industry that has been dominated by Intel and its x86-based processors.
Where once power was the key consideration for data center systems and the processors that run them, a growing emphasis is now being placed on energy efficiency. And the list of competitors, which for a long time was pretty much Advanced Micro Devices, IBM and Oracle/Sun Microsystems, now includes ARM and its roster of licensees, including Calxeda, Marvell Technologies and Applied Micro.
But despite all that—all the changing demands and workloads in the data center, all the evolving numbers of competitors—the one constant is Intel and its development and manufacturing capabilities, according to Bryant, who took over as senior vice president and general manager of Intel's Datacenter and Connected Systems Group in January. Intel products can be found in servers running whatever jobs are needed, and are rapidly expanding beyond the servers and into most data center resources, from storage systems to networking gear.
The data center will remain Intel's domain, she said.
"We have all the products in our product line to cover all the workloads," Bryant said in a recent wide-ranging interview with eWEEK in Cambridge, Mass.
For years, Intel, AMD and other chip makers would release processors, system makers would put them in their servers, and organizations would buy the servers and put them into their data centers. In recent years, with virtualization and cloud computing, big data and mobility, end users are looking for systems that can handle particular workloads. Now the focus for chip makers is creating processors designed for those workloads.
For Intel, that means offering its Xeon chips for larger, traditional data center tasks and its low-power Atom systems-on-a-chip (SoCs) for microservers for high-density cloud environments, storage and communications jobs. In addition, Intel's Xeon Phi coprocessors are aimed at highly parallel workloads found in high-performance computing (HPC) environments.
It also means more competition. Not only does Intel still compete with AMD in the traditional server space, but with the Xeon Phi coprocessors, the company now vies with GPU accelerators from AMD and Nvidia. In addition, ARM, whose low-power SoCs can be found in most smartphones and tablets, and such partners as Calxeda and Marvell Technologies see an opportunity in the microserver space, particularly once chips based on its 64-bit ARMv8 architecture start coming out next year.
AMD also will begin building ARM-based server chips in 2014. Andrew Feldman, corporate vice president and general manager of AMD's Server Business Unit, said in an interview in May that ARM could account for as much as 20 percent of the server chip market by 2016.
Intel's Bryant isn't so sure. The microserver space itself is about 10 percent of the overall server market, and when ARMv8 SoCs begin hitting the market, Intel will already be on its second-generation Atom server chip, dubbed "Avoton." In addition, Intel already offers the operating system support, middleware, developer toolkits and application support most organizations rely on. It's also expensive to create such an ecosystem, and to be able to come out with new products on a regular basis.
"Our job is to make sure we do cover the entire spectrum," Bryant said.
Intel’s Bryant: Chip Maker Expanding Reach in Data Center
ARM officials have argued that there is a strong ecosystem around the company's chip architecture, from the partners who sell chips based on the designs to the Linux distribution vendors and open-source software makers that offer software for the architecture.
With the new demands for bandwidth, processing power, energy efficiency and storage in the data center, Intel officials also are looking beyond simply creating chips for servers. The company also wants to be the power behind storage and networking systems, the latter of which represents an important growth area, Bryant said. While Intel chips are found in about 93 percent of all servers, they're in only about 10 percent of network switches and routers.
Most of the silicon in those systems is still proprietary, but the introduction of software-defined networking (SDN) is creating a "growth catalyst" for the conversion of these systems to Intel Architecture, Bryant said. In April, Intel officials outlined the company's strategy around networking, SDN and network function virtualization (NFV), including introducing "Sea Cliff Trail," a reference design for an Intel-powered open switch platform. "Sunrise Trail" is a server reference design for use in SDN deployments.
Those reference designs reflect Intel's efforts to move beyond simply building chips for systems and becoming a larger player in the data center fabric space. Another example was Intel's release in February of its own Hadoop distribution for big data environments. The Intel Hadoop offering has been used by several companies in China, including carrier China Mobile and traffic management company Bocom.
Bryant said Intel released the Hadoop distribution to make it easier and less costly for organizations to adopt big data technologies.
"Right now, [big data] is pretty complex and a lot of big data projects are expensive," she said. "Our job is to make technology less expensive and complex so people will deploy it."
Intel also is looking to leverage its massive manufacturing capabilities to grow its custom chip capabilities. The vendor created its Custom Foundry business about three years ago, enabling the company to leverage Intel Architecture IP blocks to create chips for third parties. It's a business that Intel is looking to grow, Bryant said.
Right now, the company can create chips that offer "incremental" customization through such means as adjusting the number of cores and frequency, she said. However, "our goal is to get to the point where we can offer full customization," Bryant said.