Advanced Micro Devices is considering the implementation of its upcoming low-power netbook chips based on the new Bobcat architecture in low-end servers, the company said this week.
“We’re definitely in the process of examining this as a design point,” said Donald Newell, AMD’s new server chief technology officer, in an interview. “It would be foolish not to.”
AMD will start shipping its first low-power chips based on the Bobcat architecture for devices like netbooks and ultraportable laptops later this year. The chips, code-named Ontario, combine a central processing unit and graphics processing unit into one piece of silicon.
AMD has not yet offered low-power chips as part of its server offerings, and Ontario will be AMD’s most advanced x86 low-power chip. Netbook processors like Intel’s Atom processor and Via’s Nano processor are already being used in low-end servers designed for cloud computing.
Newell was appointed the company’s server chief technology on Monday, and investigating the possible use of low-power chips in servers is part of his job to map out AMD’s future server offerings. Beyond chip level improvement, he is also looking at memory and networking improvements that could help to improve server performance. Newell formerly was an engineer at Intel, where he worked on the development of system-on-chip (SoC) and data-center technologies.
There is a growing interest in building servers with low-power chips, and the experimentation is good, Newell said. Power efficiency is a big part of server ownership as companies are looking to cut costs, and low-power chips can be useful for certain workloads, Newell said.
But before Bobcat processors are offered in servers, the company needs to analyze data like power-versus-performance benefits for specific tasks, including some that are not compute or time sensitive.
“There’s only a few papers … and there’s a lot more data to collect,” Newell said. “It really depends on a number of factors … to whether or not that’s a good design point.”
A collection of low-power processors may provide better performance-per-watt compared to the faster server chips, Newell said. But traditional server chips are more responsive and reliable, and better at running more demanding workloads.
“There’s a certain amount of computation to be done, and a certain amount of time for it to be done,” Newell said. “The large cores will get more work done in a single amount of time … and get you a better answer.”
For example, figuring out an answer to search queries is better done in server subsystems using traditional server processors than netbook chips, Newell said.
However, until all the data is collected and vetted, the company’s server strategy will continue to revolve around its Opteron line of server processors.
Current Opteron server chips come with a maximum of 12 processor cores, and AMD has already announced new 16-core server chips due to ship next year. The upcoming server chips, code-named Interlagos, will be based on the new Bulldozer architecture, which the company detailed this week at the Hot Chips conference in Stanford, California.
AMD will be entering an emerging space where the pickings are slim. Dell offers low-power servers with Via’s Nano netbook processors, and startup SeaMicro developed a breakthrough server that packs in 512 low-power Intel Atom processors on miniature motherboards the size of credit cards. SeaMicro said that Intel’s Atom processors are good at handling limited workloads, and typically provide fast responses using less power, preventing the waste of a more powerful core to provide the same results.
And after a few years of silence, Arm is also looking to push its low-power processors into servers. Companies including Marvell and Smooth-Stone have announced plans to release server chips based on Arm processors. Arm’s chips currently go into smartphones, consumer electronics and appliances.
But Newell was skeptical about Arm’s success in the server market, and may have challenges cracking into the x86 server infrastructure, which dominate data centers. Arm and x86 are different chip architectures and may not interoperate in data centers.
“It’s hard for Arm to move up in the server world, like x86 would be to move down to dishwashers,” Newell said.
AMD also is looking to mold graphics processors and separate accelerator units into its server offerings. Right now GPUs and accelerators are designed for specialist computing needs, but the company wants to build chips where all the architectural elements flawlessly work together, Newell said.