Samsung, Intel May Be on Course for Server Chip Rivalry
Samsung Electronics appears to be prepping to make a push into the server processor market, taking on x86-based chip makers Intel and Advanced Micro Devices with its ARM-designed products.
Samsung over the last couple of years has added a number of executives and engineers with server processor experience to its research and development facility in Austin, Texas. Its latest addition appears to be Pat Patla, a former vice president and general manager of AMDs Opteron server chip business who left that company last week and is now a vice president at Samsung, according to his LinkedIn profile.
That follows several other hires over the last couple of years of former AMD officials who now work for Samsungs Austin facility, fueling speculation that Samsungtraditionally the world's second-largest semiconductor maker behind Intelis gearing up to take a run at Intel and its server chip dominance.
It wouldnt surprise me, Charles King, principal analyst at Pund-IT Research, told eWEEK. It sounds like some of the hires theyve made are all deeply versed in server technology.
King also noted that some of these hires also have strong system-on-a-chip (SoC) expertise, so Samsungs motives might not be exclusively server-oriented. But its an indication that theyre at least investigating [a server chip push].
Such an effort should be expected. ARM Holdings executives have been vocal about their intention to create chip designs with the features needed for servers, including 64-bit support. In October 2011, ARM introduced its ARMv8 architecture, which includes 64-bit capabilities. However, officials said it could be 2014 before ARM-based systems hit the market.
AppliedMicro made a splash around the same time, announcing 64-bit-capable prototype chips based on ARMv8, and such vendors as Marvell Technology and Calxeda said they were working on 32-bit ARM chips that would be compatible with ARMv8.
King said that having a heavyweight like Samsung pursuing ARM-based server chips would validate the efforts being made by ARM and other manufacturers.
Keith Hawkins, who had been with AMD for more than 15 years before spending one year with Sun Microsystems, has been vice president of design at Samsungs Austin facility for two years. On his LinkedIn profile, Hawkins wrote that Samsung is continuing to expand in most design areas especially the front-end space for System, SoC and CPU architecture, verification and performance modeling. We now have our lead CPU and System Architects in place as well as our verification and performance leads. We are about to kick it in high gear with our newest design program.
In June 2011, Samsung also hired Jim Mergard after he spent 16 years with AMD. Two months later, the company hired Brad Burgess, who had worked at AMD for eight years and Intel for three years before that.
Intel executives have been somewhat dismissive about the idea of ARM-based chips gaining much traction in the server market, arguing that the chips lack key featuresincluding performance, 64-bit capabilities and software supportnecessary in the server market. In addition, Intel and AMD both have been pushing to drive up both the performance and energy efficiency of their own server processors.
However, analysts said that with the rise of cloud computing and Web 2.0 companies like Facebook and Googlewith their massive data centers processing high numbers of relatively small workloadsthere is a growing market for low-power, high-performing servers, and that ARM designs, which dominate the mobile computing space in smartphones and tablets, could find their way into such microservers.
For many applications that are more power-sensitive than performance-sensitive, ARM is going to have a real opportunity, Nathan Brookwood, an analyst with Insight 64, said in an interview with eWEEK.
That opportunity also is getting the attention of some major systems makers. Hewlett-Packard in November 2011 announced its Project Moonshot, an effort to create extremely low-power servers to run very large cloud and Web 2.0 data centers, where cooling and power costs are key concerns. As part of the initiative, HP is working with Calxeda to create systems powered by ARM-based chips.
In addition, Dell officials have said they are keeping an eye on the developments, and routinely run test systems powered by non-x86 chips, including ARM processors. However, Forrest Norrod, vice president and general manager of Dells Data Center Solutions group, in an interview with eWEEK last year questioned whether ARMs opportunity to make a mark in the server space is dwindling, saying the move is taking longer than he expected, particularly in the area of software support. Still, Norrod said, The ecosystem is not quite there yet. But its not insurmountable.
Insight 64s Brookwood said 64-bit capabilities will be important for ARM chips, but noted that many cloud computing deployments use Linux-based applications, so software support for Windows is not as important in that arena. Its much less of a big deal here than it is in more traditional data centers, he said. Instead, metrics such as power-per-wattwhere ARM has a significant advantage over x86 chipsare crucial in cloud and Web 2.0 environments.
So is server density, where SoCs with built-in fabrics will be important. AMD made a step in that direction with its acquisition in February of SeaMicro, which was making ultra-dense servers that included Intel chips and its own fabric technology. AMD will be able to leverage the SeaMicro products immediately, but it will take time for the company to tightly integrate AMD products with SeaMicros technology, Brookwood said.
Intel is a little farther behind, though, he said.
Diane Bryant, vice president and general manager of Intels Datacenter and Connected Systems Group, said after the announcement of AMDs SeaMicro deal that Intel has a very robust and compelling road map for this [microserver] market, and we partner with Dell, Supermicro, NEC and Hitachi with others to be announced soon. Weve been also collaborating extensively with HP on low-power servers and scale-out software research.
However, Brookwood noted that Intel has not been clear about its own fabric plans, and it could take a whileeven for such a well-funded and well-staffed company like Intelto develop its own chip fabric.