AMD, Dell EMC, Google, HPE, IBM, Mellanox, Micron, Nvidia, and Xilinx today are jointly announcing the launch of the nonprofit OpenCAPI Consortium, a new industry group that will promote new specifications for servers and other data center hardware. The group says that gear complying with the specs — which also use the name OpenCAPI — will achieve up to 10 times faster performance and that this sort of hardware will become available in the second half of 2017.
The server specification, which will become available before the end of this year, calls for the chip to be physically closer to the storage media. OpenCAPI-based servers, the organization says, can transfer 25 gigabits of data per second, while the current PCIe specification can transfer 16 gigabits per second.
Companies will be able to build products that meet the specs by joining the consortium or by getting a license from the group. There will be IBM POWER9-based servers, “Zaius” servers from Google and Rackspace, and Xilinx field programmable gate arrays (FPGAs) that meet the specs, according to a statement. CAPI stands for the Coherent Accelerator Processor Interface (CAPI) in POWER8-based servers.
The launch is reminiscent of the Open Compute Project that Facebook introduced in 2012 to share and collaborate on designs for its data center infrastructure. That venture has produced collaborations on the compute, storage, and networking levels.
The OpenCAPI work builds on IBM’s efforts to unite other vendors around its POWER chip architecture through the OpenPOWER Foundation. Google, IBM, Mellanox, and Nvidia, among others, were on board for the establishment of the OpenPOWER Consortium in 2013.
The organization’s board includes one director from each of the founding companies. The board takes input from a technical steering committee, whose members are elected representatives from companies that are members of the organization.
For more detail, see the organization’s website.