Advanced Micro Devices is betting on its upcoming “Zen” core architecture to put it back in play in the data center and in PCs against larger rival Intel.
Company officials introduced Zen in May 2015, outlining an architecture that will support simultaneous multi-threading (SMT)—a technology similar to Intel’s Hyper-Threading—and DDR4 memory, and will feature a FinFET transistor design for the 14-nanometer chips. It will provide a 40 percent improvement in instructions-per-clock over current AMD processors.
The Zen chips will first appear in high-end desktop PCs later this year and then afterward in servers, with full revenue ramp beginning in 2017. AMD executives have talked about some aspects of the architecture, but haven’t elaborated beyond the basics.
However, a presentation by a researcher at CERN at a recent IT forum unveiled some more information about AMD’s plans for Zen, in particular that upcoming server chips will feature as many as 32 cores—packaged across two 16-core modules—and will support PCIe gen 3.0 interconnect and will leverage as many as eight channels of DDR4 memory. CERN is the Swiss scientific institution that built and operates the Large Hadron Collider.
The high number of cores and other features of Zen—including the FinFET transistor design, which will help drive performance and power efficiency—show the effort AMD is putting behind the core design, which officials have said was two years in the making.
CEO Lisa Su and other AMD executives have laid out an aggressive, multi-pronged strategy to bring the company back to sustainable profitability later this year and to compete better with Intel on multiple fronts, including high-end PCs and servers. AMD also is targeting other areas, including gaming and immersive computing.
The company last year made strides on multiple fronts, including its GPUs and PC processors. The company released its “Carrizo” processors for notebooks and other PCs, and expanded its graphics offerings as well as creating a GPU business unit, the Radeon Technologies Group.
AMD also wants to challenge Intel’s dominance in the data center, where the larger vendor owns more than 98 percent of the market. About 10 years ago, AMD, on the strength of its Opteron processors, was able to claim more than a quarter of the server market, but lost most of that ground due to Intel’s innovations in its Xeon processors and missteps by AMD.
Analysts have said that OEMs and end users are looking for an alternative source of processors to Intel to help drive innovation, reduce prices and protect themselves if there ever is a problem with the Intel supply chain. AMD is aiming to be that alternative on two fronts—its x86 Zen chips and its ARM-based “Seattle” processors. Other ARM chip-making partners—including Qualcomm, Applied Micro and Cavium—also are making systems-on-a-chip (SoC) for the data center, while the IBM-led OpenPower Foundation expects to become that second-source alternative. Su told eWEEK in December that Zen shows that the data center is “not an all-[Intel] Xeon play.”
The 32 cores and eight-channel DDR4 memory interface are about double what Intel offers now in its high-end Xeon processors, and the move to 14nm will put it on par with Intel. Currently, AMD’s server chips are built on a 28nm manufacturing process.