IBM's new Power8 doubles performance of Watson chip

IDG News Service - IBM for the first time revealed details on Monday of its 12-core Power8 chip, which is twice as fast as the Power7 chip used in the Watson supercomputer.

To make the chip faster, IBM has turned to a more advanced manufacturing process, increased the clock speed and added more cache memory, but perhaps the biggest change heralded by the Power8 cannot be found in the specifications.

After years of restricting Power processors to its servers, IBM is throwing open the gates and will be licensing Power8 to third-party chip and component makers.

IBM has moved away from some proprietary board-level technologies with Power8, and has included a connector so third-party graphics processors and other components can be easily linked to the chip.IBM recently announced it would open up the Power chip intellectual property and license it to third parties including Google. The IP is being opened up as part of a development alliance called OpenPower Consortium, and one of the members is Nvidia, which is expected to develop a graphics processor that connects to the Power8 processor. Tyan is building a server based on the Power8 chip.

The previous Power7 chip was perhaps best known for its place in the Watson supercomputer, which famously competed against humans on the U.S. TV quiz show "Jeopardy" -- and won.

Watson also found use in areas like health care and the financial sector. Beyond IBM's traditional market of Unix servers, the Power8 chip is also designed for areas like cloud and big data."Big data is something that is driving [performance]," said Jeff Stuecheli, chief nest architect of IBM Power Systems in the Systems and Technology Development Group. "You have big data, you need big performance."The Power8 chip will support up to 1TB of DRAM in initial server configurations and will offer 230GB per second of sustained memory bandwidth, Stuecheli said.

The external components will be connected via the CAPI (Coherence Attach Processor Interface) port. The CAPI port interfaces with the PCI-Express slot for external components such as GPUs or FPGAs (field programmable gate arrays) to communicate with the chip.The chip is made using the 22-nanometer process. Power8 has a large cache memory, including 512KB of cache per core, 96MB of on-chip shared L3 and 128MB off-chip L4 cache. The L4 cache was removed in Power7, but it comes back with Power8. Each processor core will also support eight threads, giving the chip the ability to run 96 threads simultaneously.IBM wants to share the Power8 technology with a larger ecosystem, and is also working on an open software stack for the processor, Steucheli said. A lot of the tools will help develop high-performance applications.

Stuecheli declined to say when the Power8 chip would be released.

Agam Shah covers PCs, tablets, servers, chips and semiconductors for IDG News Service. Follow Agam on Twitter at @agamsh. Agam's e-mail address is agam_shah@idg.com

Reprinted with permission from IDG.net. Story copyright 2012 International Data Group. All rights reserved.

View the original article here

IBM's Watson could get even smarter with Power8 chip

IBM’s “Watson” technology, released in 2009, put a face on what the massive processing power of a mainframe could mean in the real world. Now, with IBM’s new Power8 chip, Watson coud grow even smarter.

IBM launched Watson on its Power7 chip, according to Jeff Steucheli, with IBM’s Power team, during a presentation Monday at the Hot Chips conference Monday at Stanford University. But according to IBM’s own internal metrics, the Power8 is between two and three times faster than the Power7, launched in 2010. (An intermediary chip, the Power7+, was launched in 2012.)

IBM maintains a specialized line of chips, known as the Power line. Most of the supercomputing and server world have moved to Intel’s Xeon processor, however, making Watson both an important technical as well as marketing tool to show off the power of Power.

Watson, of course, trounced the top players in the TV quiz show, Jeopardy. Watson was later adapted for front-end customer-service applications, financial analysis, and even has its sights set on smartphones.

Put simply, Watson was little more than a specialized database connected to a front-end interface that could understand speech in a natural-language context. With Power8, IBM has more than doubled the sustained memory bandwidth from the Power7 and Power7+, to 230 GB/s, as well as I/O speed, to 48 GB/s. Put another way, Watson’s ability to look up and respond to information has more than doubled as well.

“You can imagine if you have 3x the performance of a Power7, you can do some very interesting things,” Steucheli said.

Recently, IBM announced its OpenPower initiative, where it will license the Power chips and co-develop an ecosystem around the Power architecture with companies like Google, Nvidia, and Mellanox. Up until now, IBM primarily used the Power design in its own servers. This new initiative makes it possible for cloud services and their technology providers to redesign the chips and circuit boards where computing is done, optimizing the interactions of microprocessors, memory, networking, data storage and other components, IBM executives said.

“Watson wasn’t a traditional workload for us,” Steucheli said. “We’d like to find more of these opportunities.”

IBM hasn’t said when the Power8 will ship. “But I have a processor here,” Steucheli said Monday. “And we have a lab full of these things.”

View the original article here