AMD’s Future Includes ‘Zen,’ ARM, High-Bandwidth Memory

by Jeffrey Burt

“We don’t need to do everything,” Su told analysts and journalists. “We need to pick the things we can do very well.”

Such as in the gaming space, immersive platform segment and the data center, according to company executives.

Including low-end PCs, smartphones, tablets and endpoints for the Internet of things.

AMD in April shed its SeaMicro microserver business, and Su said during the meeting that the company was ending its SkyBridge project of developing socket compatibility between x86 and ARM cores.

Central to AMD’s short-term goals will be the Zen x86 CPU core design that’s been in the works for two years. Set to be released in 2016, Zen will feature simultaneous multithreading (SMT), support for DDR4 memory, a FinFET transistor design and a performance boost of 40 percent over current chips.

AMD will continue pushing forward on its plans to become the top vendor of ARM-based server chips, starting with the long-awaited Opteron A1100 “Seattle” processor, which will come later this year. Forrest Norrod, senior vice president and general manager of the company’s Enterprise, Embedded and Semi-Custom Unit, said customers still want a choice other than Intel in the data center.

The vendor’s custom server and embedded system on a chip (SoC)—which will be built using an ARM architecture license—will launch in 2017.

The company’s accelerated-processing units—which feature integrated CPUs and GPUs on the same piece of silicon and have been in PCs for several years—will come to servers in the future, according to Mark Papermaster, CTO and senior vice president of technology and engineering.

AMD will begin putting high-bandwidth memory into its next-generation Radeon GPUs. The technology essentially stacks memory chips for greater performance, power efficiency and density, and offers three times the performance-per-watt of the current GDDR5 and a 50 percent increase in energy efficiency, AMD says.

The vendor’s network-on-a-chip technology design method takes advantage of reusable IP building blocks to maximize efficiency, which in turn should lower cost and time-to-market for AMD products, according to Papermaster.