SAN FRANCISCO—Fifteen months after announcing plans to roll out an entirely new x86 processor microarchitecture that would scale from high-end servers down to notebook PCs, executives with Advanced Micro Devices laid the groundwork for chips that will feature the new “Zen” core, which they expect will become the cornerstone of a strategy to win back business and consumer PC users and become a significant player again the data center.
During an event here Aug. 17 before a group of journalists and analysts, AMD President and CEO Lisa Su and Mark Papermaster, senior vice president and CTO, demonstrated working processors aimed at high-end desktop PCs and servers that showed significant improvements in performance and power efficiency over company’s current generations of processors. The chips, they said, will compete favorably with Intel’s portfolio of chips.
They also provided a clearer road map of the Zen-based chips’ roll-out, noting that desktops powered by the upcoming “Summit Ridge” chips are planned to begin shipping in the first quarter 2017 although some may hit the market late this year, Su said. The new chips will use the same AM4 socket as the company’s latest seventh-generation A-Series chips announced this summer, which include DDR4 memory, PCIe Gen 3 interconnect and SATA Express.
The Zen-based server chip—dubbed “Naples”—is scheduled to begin shipping in systems during the second quarter next year, followed by Zen-based notebooks in the second half of 2017. CPUs and accelerated-processing units (APUs)—which include both the CPU and GPU on the same piece of silicon—are slated to begin shipping after that.
The demonstrations of the chips and the new timeline information comes more than a year after Su, Papermaster and other AMD executives outlined their plans for reviving the flagging company and returning it to sustainable profitability. Those plans, revealed in May 2015, included focusing the chip maker’s efforts on a particular set of markets, including high-end client systems, servers in the data center and immersive computing environments, such as virtual reality (VR) and gaming.
Over the past 12-plus months, AMD has made a number of moves to meet those goals. The company has put a focus on its graphics technologies, creating the Radeon Technology Group business unit and, more recently, rolling out new GPUs based on its latest Polaris architecture. In addition, AMD in June launched its seventh-generation A-Series chips that officials said would challenge Intel’s Core processors in the PC space, and has entered into joint venture deals to increase its presence in the huge Chinese market.
AMD’s semicustom chips also now power both Microsoft’s Xbox and Sony’s PlayStation game consoles, and more semicustom chip deals are on the way, Su said.
Zen Chips at Heart of AMD’s Plans to Revive Profits
At the center of AMD’s plans was the Zen, a new architecture built from the ground up to help the company meet the enterprise demands for greater scalability, performance, power efficiency and affordability brought on by the cloud, VR and augmented reality, data analytics, big data and other trends. The goal of Zen was to remake the x86 architecture for a more modern era by driving up the performance of chips while keeping a lid on power consumption, Papermaster said during the event.
Last year, he set the goal of hitting a 40 percent improvement in instructions-per-clock performance over AMD’s current chip—a goal that he showed off during the demonstration of an eight-core, 16-thread Summit Ridge chip running the multithreaded Blender rendering software. In addition, in a head-to-head comparison with a similarly configured “Broadwell-E” processor from Intel, the Summit Ridge chip delivered comparable performance and power efficiency.
The demonstration of a 32-core, 64-thread Naples server system-on-a-chip (SoC) also showed significant performance and efficiency gains.
In an interview with eWEEK, Su said AMD “made a big bet” by spending the time and resources to create a new microarchitecture, but it was a “high-risk, high-reward” bet that needed to be made for AMD to push its way back into the picture for both PCs and servers. With Zen hitting the performance goals that were set more than a year ago and the expected launch of the first chips just months away, the bet is paying off, she said.
“I would agree that this is the biggest product launch [for AMD] in 10 years,” Su said. “We’re exactly where we thought we would be.”
Industry analysts were impressed with what they saw during the event, though they were cautious when discussing AMD’s future.
“Although they still have a ways to go … I can see them becoming competitive once again in desktops and servers,” Nathan Brookwood, principal analyst with Insight 64, told eWEEK.
Similarly, Patrick Moorhead, principal analyst with Moor Insights and Strategy, said he liked what he saw, though he also was wary.
“If they can deliver on this, it will be the biggest thing they’ve delivered in 10 years,” Moorhead said.
He was referring to more than a decade ago, when AMD surprised Intel with the delivery of Opteron, the first 64-bit x86 server chip that offered greater performance and power efficiency than anything its larger rival could offer. The market embraced Opteron, and for a while AMD was able to grow is share of the x86 server market to well over 20 percent. However, between Intel marshaling its considerable resources to catch up and missteps by AMD, that market share dwindled, and now Intel chips run more than 95 of all data center servers.
AMD Puts Spotlight on Upcoming Zen Processors
However, Brookwood and other analysts said that server OEMs and businesses are looking for an alternative to Intel to help drive innovation and keep costs down, and AMD, with its Zen microarchitecture, could be that option. Other chip makers—including ARM and partners Qualcomm, Applied Micro and Cavium, not to mention IBM and its OpenPower effort—-also are vying to be that Intel alternative.
Where AMD Has a Leg Up Over Rivals
However, Brookwood noted that AMD offers advantages that ARM and IBM can’t match. In particular, AMD’s technology is based on the same x86 architecture as Intel, which would make it much easier for enterprises to shift their workloads to AMD from Intel. Turning to ARM or IBM would require embracing another architecture and all the work it would take to port their software, he said.
Su also noted that unlike ARM and its various partners, AMD has a long history of developing silicon for the server market.
“x86 is the dominant architecture when talking about the data center and when talking about PCs,” she said. “Experience is important. The number of players who can do high-performance CPUs is few.”
She said during the event that “Zen is absolutely critical to our re-entrance into the data center. There needs to be competition in the data center.”
Papermaster said that in building Zen, AMD engineers looked at everything to help drive performance and power efficiency increases. Among the key features leading to the gains are simultaneous multithreading—a technology similar to Intel’s Hyper-Threading—for improved performance, a new cache hierarchy, improved branch prediction and the use of FinFET technology for better performance and efficiency.
“Performance, throughput and efficiency,” he said, noting that AMD officials will talk more about the Zen microarchitecture at the Hot Chips 2016 show, scheduled for Aug. 21-23 in Cupertino, California. “That’s what the team put together. Those were the levers” pulled to get the performance gains.
It was important for AMD to make the effort to develop Zen as it looked to make its way back into servers and to regain its footing in PCs, especially given the dominance of Intel in both markets, Su and Papermaster said.
“We have to be faster, more flexible [and] creative because we’re the smaller player,” Papermaster said.
In response to a question during the event regarding Intel, Su said that “we will always be paranoid about our competitor, but we are playing our game.”
She told eWEEK that the disruptions in the market caused by the cloud and other trends are creating opportunities for AMD to grab share from Intel.
“The market is changing,” Su said. “It is different from the way it was five years ago, even 10 years ago.”
Even as AMD is gearing up to finally get Zen into the market, Su and Papermaster cautioned against expectations of rapid results. The CEO told eWEEK that while adoption can happen relatively quickly in the desktop PC space, the ramp up of servers in the enterprise space tends to take longer. OEMs need to qualify the chip for their systems, and then end users need to go through a test and evaluation phase before deciding to adopt the technology. Zen—particularly in the enterprise—is a long-term strategy, she said.
The company already is looking down the road, Papermaster said, with engineers already working on the follow-on to Zen, called Zen+.
What was noticeably absent from the presentation was any talk about AMD’s plans for ARM-based server chips. The company had been looking at ARM-based SoCs as complements to its x86 offerings—and released its first Opteron-A chip earlier this year—but earlier this month reports surfaced that the company had put the ARM idea on the back burner to focus more of its efforts and resources on Zen and the x86 market.