Advanced Micro Devices officials are laying out a server chip road map that includes the company’s first ARM-based processor named “Seattle” that will begin appearing in the second half of 2014.
Seattle is part of a larger effort by AMD to build Opteron server processors that meet the demands for greater system performance and energy efficiency from organizations that are building cloud environments and dealing with such trends as big data and mobility.
AMD officials on June 18 unveiled upcoming x86-based server chips—dubbed “Berlin” and “Warsaw”—that will aim for greater rack density and the two- and four-socket server sets, respectively. But it’s Seattle that will get a lot of the attention.
ARM designs chips and then licenses those designs to a wide range of vendors, from Samsung and Qualcomm to Nvidia and Texas Instruments. ARM-designed systems-on-a-chip (SoCs) dominate the mobile device space, including smartphones and tablets, but officials now believe their low-power designs could fit well in some hyperscale servers that demand greater energy efficiency.
AMD officials last year said the company would begin manufacturing ARM-based server SoCs starting in 2014, part of the company’s ambidextrous strategy of offering customers chips based on whatever architecture they demand. In a recent interview with eWEEK, Andrew Feldman, corporate vice president and general manager of AMD’s Server Business Unit, said that ARM will begin seeing tremendous traction in the server space, holding as much as 20 percent of the server chip space by 2016.
And AMD will be the dominant supplier of ARM-based server SoCs, Feldman said. There already are chip makers—Calxeda, Marvell Technologies and Applied Micro—that offer ARM server chips. However, those vendors don’t have AMD’s history of making processors, the massive amounts of IP to draw on, or the strong relationships with OEMs and original design manufacturers, Feldman said. They also don’t have the experience or resources to churn out upgraded chips every 12 to 18 months.
“The CPU business is no place for small companies,” Feldman said. “It’s just too expensive.”
ARM-based server chips will begin to see true momentum after ARM releases its upcoming ARMv8 designs, which will include such key server features as 64-bit capabilities and greater virtualization support, he said. AMD will begin with Seattle, an SoC that company officials said will offer two to four times the performance of AMD’s new x86-based Opteron-X chips.
Seattle initially will come in eight-core versions, and then 16 cores, with up to 64GB of DRAM support. The highly energy-efficient chips will be based on ARM’s Cortex-A57 cores that will hit speeds of 2GHz or more, and will include integrated 10 Gigabit Ethernet. They also will feature AMD’s Freedom Fabric—acquired last year when the company bought microserver maker SeaMicro—integrated on the SoCs.
AMD Unveils Details of First AMD-Based Server SoCs
The Seattle chips will be aimed at smaller servers—including AMD’s own SeaMicro systems—many of which are found in cloud and Web hosting environments running massive numbers of smaller workloads. AMD will begin sampling the Seattle SoCs in the first quarter of 2014.
Some top-tier server makers are beginning to design systems that run on ARM chips, looking to meet the growing demands from organizations like Facebook and Google, who run huge data centers and see density and energy efficiency as keys. Hewlett-Packard and Dell already are working with ARM chip providers.
Intel is looking to its x86-based low-power Atom platform to push back at AMD and other ARM-based server chip makers in the microserver space. Intel already has its “Centerton” SoC on the market, and its upcoming “Avoton” SoC will offer greater performance and power efficiency when it comes to market later this year. Intel officials have argued that Atom gives programmers a familiar x86 development environment and tools, and noted that by the time the first 64-bit ARM chips come out next year, Intel already will be on the second generation of its microserver processors.
AMD’s Feldman said the ecosystem that has cropped up around ARM’s designs has fueled faster innovation than in the x86 space.
“[Server] customers are eager to see the vibrant ecosystem they’ve seen on the client side,” he said.
AMD also is taking steps to make programming between x86 and ARM easier. For example, it helped found the Heterogeneous System Architecture Foundation, whose goal is to create a standards-based ecosystem that will ease the porting of applications across multiple architectures.
Even as they begin to lay out plans for their Seattle chips, AMD officials are not abandoning x86. Berlin, which will be released in the first half of 2014, will be available as either a CPU or an accelerated processing unit (APU), with integrated graphics. The chip’s four cores will be based on the upcoming “Steamroller” architecture that officials said will offer eight times the gigaflops-per-watt performance of the vendor’s current Opteron 6386SE chip.
It also will be the first APU built on AMD’s Heterogeneous System Architecture, which will give it uniform memory access for the CPU and GPU.
In the first quarter next year, AMD will release Warsaw for two- and four-socket servers. The chips will targeted at traditional enterprise workloads.