During today’s Open Compute Summit keynote, AMDs Andrew Feldman, corporate vice president and general manager at AMD, introduced the Opteron A-series server SoC that is codenamed Seattle. Before coming to the meat of the matter, Andrew talked a lot about a huge market opportunity of connecting the remaining 2/3 of the worlds population to the Internet and increasing the number of smartphone users, that currently hovers just short 1 billion. However, most of these smartphones are pretty dumb unless connected to datacenters and this is where AMD comes in.
Andrew also pointed out that at this point AMD is not particularly interested into providing products for the Internet of Things. However, this recent trend will put new demand on datacenters and here is where AMD wants to capitalize. This seems to be a sound move for a company that reduced their operating profile to a lower level starting at the end of 2012. This way, AMD can still benefit from the Internet of Things trend without leaving their traditional fields of expertise.
Before going into specifics Mr. Feldman also explained the rationale behind the move to offer ARM-based servers. Due to the sheer amount of ARM CPUs shipped compared to x86 server CPUs it’s needless to say this is a big growth segment. Mr. Feldman joked that he couldn’t show up at his customers without being asked when they would be able to test their ARM parts. He concluded that he thinks the ecosystem seems to have aligned behind ARM and AMD wants to be part of that success.
AMD Announces the first 64-bit ARM Server Processor
They claim it will feature server-class IP blocks none of the ARM competitors has.
2-4X perf over Opteron X-series . Unlike the announcement of Seattle last year, this time AMD provided detailed specifications and also named the CPU. The Opteron A-series A1100 series SoC will feature 4 or 8 Cortex A57 cores clocked at or above 2GHz with up to 128GB of DRAM support. DDR3 and DDR4 memory technologies with ECC are supported with up to 1866MHz frequency. The cores feature up to 4MB of shared L2 cache in total (1MB per pair of cores) and a shared L3 cache of 8MB across all cores.
In terms of I/O the SoC offers 8 lanes of PCIe 3.0 connectivity, 8 SATA ports and two 10GbE interfaces for fast network connectivity. The SoC also comes with ARM TrustZone support and features co-processors for encryption and compression acceleration. The chip comes in at an expected 25W TDP. AMD touts their experience as an established server vendor as one of their key advantages over competitors.
When AMD emphasizes their server experience and intellectual property, they really mean technologies which are the bread and butter of todays commodity x86 servers but are fairly uncommon in the ARM ecosystem yet. These are technologies such as ECC memory support, high performance memory controllers, fabric technology used to interconnect a large number of servers or storage subsystems that can drive a large number of storage devices such as hard disks and SSDs. AMD now wants to offer these ingredients in both x86 and ARM flavors going forward.
In line with what the company announced last year, Seattle will begin to sample in a few weeks from now, in March specifically. AMD demonstrated a working development board at the OpenCompute Summit, as well as a server on a card for the common slot specification. AMD contributes the latter design to the OpenCompute Project.
The Impact of ARM Servers on AMD
After the announcement at the OpenCompute Summit we had the opportunity to discuss the impact of the new ARM server processors on AMDs business with Andrew Feldman. The most interesting aspect lies in the significance of the x86 and ARM technologies for AMD going forward. As current forecasts show, by 2019 CPUs based on an ARM architecture will comprise 25% of the server market, with the remaining 75% still being x86 based. However at AMD, Mr. Feldman actually expects a fifty-fifty share at AMD by that time. He admitted that in terms of revenue the bigger share would still be x86 due to the ARM parts being less expensive.
When pressed further by us about gross margin expectations the investor relations advisors stopped Andrew from being too specific, but he explained that the company is committed to keep the margins as is on the x86 side. About the margins for ARM server SoCs it is still too early to talk as the market is yet to develop and shape over the course of the next years.
Given the strong expectations of customer uptake we asked whether he believes ARM to be superior to x86. Without directly answering our question, Mr. Feldman explained that the company wants to offer choice and will continue to pursue customers in the x86 segment. From now on we should expect AMD to be committed to both technologies. AMD wants to regain some share in x86 and lead the market in ARM, as Andrew puts it. He expects a more diverse market for server processors with different capabilities to develop, including different instruction sets, big cores, small cores, GPUs, application specific accelerators and other capabilities that can be better tailored to specific needs.
The typical application areas for the ARM-based SoCs AMD recognized demand for are web servers, storage and memory caching. The company expects these areas to grow over time. In the coming years Mr. Feldman expects cutting edge companies the primary customer of such SoCs, with more traditional customers following later on. The current software environment centers primarily around open source software in the Linux ecosystem, such as the classic LAMP stack (Linux, Apache, MySQL, PHP). Furthermore we should expect to see ARM-based cloud offerings of all the major cloud vendors in the near future.
The advantage in cloud computing is that in many application scenarios the instruction set of the CPU driving the services doesn’t matter at all. In this segment important dimensions to optimize are performance per dollar, performance per watt and a scalable architecture. As Andrew aptly explained in his keynote talk, "smaller, lower power, lower cost CPUs have always won."