Seven technology companies have announced that they will band together to form the CCIX Accelerator Consortium in an attempt to interconnect their CPUs, accelerators and networks.
Advanced Micro Devices (AMD), ARM Holdings, Huawei Technologies, IBM, Mellanox, Qualcomm and Xilinx reached an agreement to develop and implement an interconnect that will allow each of the companies' CPUs and accelerators to communicate with one another while sharing a system's main memory.
The CCIX Accelerator Consortium has a long road ahead of it to make its shared vision a reality. It is already quite difficult to build a cache coherent interface between two or four of the same type of chips such as CPUs. The group however plans to take this idea a step further by building an interface that would allow CPUs, FPGAs, GPUs and network chips to work together while sharing the same cache, which will present plenty of comparability issues and difficulties along the way.
If these seven hardware vendors are able to accomplish their goal the potential benefits would be substantial. Plug and play computing, network acceleration and better performance than what is currently available through PCIe could all be gained by an interconnect.
The accelerators that exist today typically use PCIe (PCI Express) as a means of connecting to processors. This technology has been around since 2003 and was never intended for high bandwidth, low latency connectivity that is needed between processors. By having all of the components and accelerators exist on a shared memory bus, they would all have access to high speed memory ensuring that there would be no bottleneck between these different parts of a system.
The CCIX consortium will deliver more information in the coming months and hopefully it will shortly begin work on its goal of delivering an interconnect between the CPUs, accelerators and network chips of its members.
Image Credit: HelloRF Zcool / Shutterstock