IBM hopes to spawn third-party Power servers by opening up its 12-core Power8 chip design to licensees, and now the company has to convince component makers to make parts for the servers.
With Power8, IBM is providing hooks so component makers can easily plug their parts into the new breed of servers, said Bill Starke, Power chip architect, in an interview during the Hot Chips conference this week in Stanford, California. IBM presented technical details of the chip at the show.
IBM now has to help build a secondary third-party component industry to ensure parts are readily available for servers. The company has made some interface changes to its Power8 chip design to ensure outside components can communicate with the CPU and other processing units.
IBM in early August made the surprise announcement that it would license its Power architecture for the first time to third parties such as server and component makers as part of a development alliance called OpenPower. Power was previously relegated to homegrown IBM servers and custom chips, and the company now hopes to see more Power systems in the market.
OpenPower started with a handful of members such as Google, which has hinted at the possibility of building a server. Another licensee is system builder Tyan, which will be the first company outside of IBM to offer a Power server. Other licensees include graphics card maker Nvidia and Mellanox, which is best known for products based on the InfiniBand networking interconnect, which is widely used in supercomputing.
Another surprise in Power8 is IBM's move to the industry standard PCI-Express 3.0 protocol, which the company admitted was much faster than its proprietary interconnect used in Power7 and Power 7+ servers. Oracle has also moved over to the PCI-Express 3.0 protocol with the Sparc M6, which was also detailed at the Hot Chips show.
Also new is the CAPI (Coherence Attach Processor Interface) interface so a wider range of components can work with the processor. CAPI helps connect third-party hardware such as graphics cards, storage devices, and custom chips such as FPGA (field-programmable gate arrays) and ASICs (application-specific integrated circuits). CAPI should also make it easier for server makers to build off-the-shelf Power systems, much like barebones servers today.
The CAPI interface sits on top of the PCI-Express bus, and is necessary for chips that wouldn't otherwise be able to attach to PCIe slots, such as FPGAs. Traditionally FPGAs, which are reprogrammable chips, were installed directly on the motherboards of Power servers.
Power8 is mostly targeted at high-performance supercomputers, and CAPI can support chips that require high bandwidth. It also allows for highly parallel processing, in which third-party components can harness the speed of Power8 to execute tasks faster, Starke said.
Power8 will offer two to three times more performance than the previous Power7 and Power7+ processors, IBM said. Power8 is made using the 22-nanometer process, which is also helps deliver more performance and power savings.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.