When Advanced Micro Devices launched its first Opteron server processor in 2003, it sent a jolt through the server market.
At the time, larger rival Intel was pushing its Itanium platform for the growing numbers of 64-bit workloads coming into the marketplace, moving away from the x86 architecture that formed the foundation of its Xeon processors.
AMD officials argued that there was no reason to abandon the x86 architecture, and with Opteron introduced the first 64-bit x86 processor. It could run both 64- and 32-bit applications, and was embraced by both server OEMs and end users, persuading Intel to bring 64-bit capabilities to its Xeon processors and make Itanium a niche high-end technology. Within a few years, AMD’s share of the x86 server chip space grew to more than 26 percent.
AMD’s market share now stands at about 1.5 percent—Intel’s is more than 98 percent—due to technological and market missteps, focus on such aspects as power efficiency and core counts, and most recently an effort to develop low-power ARM-based server chips.
“AMD really hasn’t been focused on instructions-per-clock as it relates to their core,” Patrick Moorhead, principal analyst at Moor Insights and Strategy, told eWEEK. “They haven’t been investing in their core IP.”
That’s about to change, according to President and CEO Lisa Su and other company executives. Speaking to financial analysts May 6 in New York City, Su said the data center will be one of the three legs of the stool—the others being gaming and immersive computing—that will bring AMD not only back to profitability, but to sustained profitability. And the primary focus of that effort will be the x86 server space. The company is eschewing many low-margin businesses—such as low-cost PCs—to invest in areas that require high-performance computing. The x86 server space—particularly with the growing compute-intensive workloads like big data analytics—is one of those areas.
AMD’s data center focus will include not only servers, but also workstations, networking and storage. However, Su and other executives spent much of their time talking about plans for the x86 server market. The data center “is probably the biggest single bet we’re making today,” she said.
Forrest Norrod, senior vice president and general manager of the company’s Enterprise, Embedded and Semi-Custom Unit, said the growing need for choice in the x86 server space was important not only to drive competition, but also for improving the economics and innovation in the space.
“So AMD is getting back into the x86 server market and providing that choice,” said Norrod, who came to AMD several months ago after serving as head of Dell’s server business.
There are a number of moving parts to AMD’s plans. The company is developing new x86 Opterons based on the upcoming “Zen” core design that will first appear in PC chips while at the same time changing the timelines for its ARM-based server systems-on-a-chip (SoCs). There also are plans to create a high-performance server accelerated-processing unit (APU) that will include the CPU and GPU on the same die. Norrod and Su said they expect to start seeing gains in the data center space in 2016 and ramping into 2017.
At the same time, AMD is shedding efforts that didn’t fit in with plans. The company announced last month that it was ditching its SeaMicro microserver business, which it bought in 2012 for $334 million. The dense microserver market was not growing as quickly as expected, and officials wanted to get out of the systems business, Su said. In addition, the company has ended its SkyBridge project for developing pin-compatible x86 and ARM chips, with the CEO saying there was little customer demand for it.
Instead, the company in the second half of the year will launch its long-awaited ARM-based Opteron A1100A “Seattle” SoC, which leverages ARM’s Cortex-A57 cores. The chip was first sampled in the first half of 2014, but Su and Norrod said there were adjustments the company needed to make to the design based on feedback from the sampling before releasing the processor.
AMD Aims to Reinvigorate x86 Server Business
However, much of the focus now is on the upcoming x86 Opterons that will be based on Zen. The new core design will come with an array of new features, from its support for DDR4 memory to its embrace of simultaneous multi-threading (SMT) rather than AMD’s Cluster-based Multi-Threading. It also will feature a FinFET transistor design for the 14-nanometer chips and will include a new high-bandwidth, low-latency cache system. The design promises as much as a 40 percent performance improvement over current chips.
New Opterons featuring the design are expected to begin appearing in 2016, according to the company.
Moorhead, of Moor Insights and Strategy, said such performance gains will be impressive if the technology comes through.
“That’s a really good sign,” he said, adding a cautionary note. “It doesn’t mean market success. It positions them for possible market success.”
It also helps the company move up the competitive ladder. Where current Opterons based on such architectures as “Bulldozer” were positioned to compete with Intel’s low-end Xeon E3 chips, Zen will enable AMD to compete in the midrange against Xeon E5 chips, he said.
While putting more effort into the x86 space, AMD isn’t abandoning ARM. The company is looking to launch its K12 server SoC—created through an ARM architecture license—in 2017.
Su said AMD will have to be careful how it competes with Intel, which is much larger and has more resources in such areas as money, manufacturing and people. That challenge was on display during the week—the day before AMD’s analyst meeting, Intel launched its powerful Xeon E7 v3 high-end server processors. Norrod said there will be opportunities to focus on doing “things that Intel doesn’t want to do, or things Intel isn’t doing,” harkening back to the days when AMD was the first to bring 64-bit computing to x86 chips, or integrating the memory controller onto the chip.
Moorhead said the industry’s desire for a second supplier of x86 chips not only helps with pricing and innovation, but protects customers against any sourcing problems. Intel also would probably welcome a stronger AMD, both Moorhead and Norrod said. The ARM architecture is still out there as a possible alternative, and IBM is making strides at the high end with its Power8 chips, Moorhead said. A strong AMD would mean that more workloads would stay in the x86 camp, and workloads lost to AMD could find their way back to Intel, Norrod said. If a business moves off x86 and onto ARM or Power, they probably wouldn’t return to x86, he told analysts and journalists after the meeting.
That brings into question what will be the alternative to x86, Moorhead said. The industry wants to have two architectures to choose from.
“We know one of them is x86,” he said. “Then it becomes, will it be ARM or will it be Power?”
ARM’s server future remains unclear. The company now offers a 64-bit architecture, and a number of chip makers—not only AMD, but Qualcomm and Applied Micro—are developing products. However, at this point, Applied Micro and Cavium are the only ones to have chips on the market. There has been some interest—PayPal is using servers based on Applied Micro’s X-Gene chips, while the University of Utah’s CloudLab project is running on Moonshot servers from Hewlett-Packard powered by the Applied Micro SoCs—but long-range success is uncertain.
IBM continues to build out its Power portfolio, while also expanding the reach of the architecture through its OpenPower efforts.
AMD’s Norrod, dating back to his Dell days, has taken a cautious approach to ARM servers, questioning whether the demand is there beyond the massive Web-scale businesses and other niche markets. Still, he said following the AMD meeting that ARM’s architecture also is playing a role in other areas of the data center, including storage and networking.