With the rise of high-end hardware, the bottlenecking in modern systems has now shifted to the connections between individual components. Two open technologies, OpenCAPI and Gen-Z, have been designed to help.Stefan Bordel and Jens Stark
Computers are expected to do more, faster: processors, RAM and hard drives are getting quicker all the time. The bottleneck has now become the connections between those components. To ensure that future servers and high-performance computers can handle rising data throughput, the tech industry is now working with OpenCAPI and Gen-Z, two technologies intended to help resolve the issue.
OpenCAPI is an open interface originally created by IBM and now maintained by a consortium composed of IBM, AMD, Google, Micron and Mellanox, among others. OpenCAPI is intended to connect CPU, GPU, RAM and memory using methods similar to PCI-Express 3.0, only ten times faster. That represents speeds of 150 gigabytes per second.
Future generations of computers featuring the interface will thus be well equipped for data-intensive tasks such as virtual reality, augmented reality, machine learning and complex scientific calculations. Machines running the technology will also be able to accommodate quicker components such as high-speed 3D Xpoint memory. OpenCAPI will first see use in high-end servers and supercomputers. The technology is intended to eventually make its way to desktop computers as well. The first OpenCAPI ports will be installed in IBM's Power9 servers, slated to launch next year. In the meantime, AMD has announced plans to integrate the technology into its Radeon GPUs.
Gen-Z technology, a new transfer standard, is intended to optimize networking between various devices and provide higher data transfer rates. Gen-Z will likely be of most benefit to computer centers. The technology is also maintained by a consortium featuring the likes of AMD, ARM, Broadcom, Cray, Dell EMC, Hewlett Packard Enterprise, Huawei and Samsung.
Gen-Z allows for computer centers to install individual server racks stocked solely with processors; the associated RAM actually housed within and connected to other elements. This is intended to boost efficiency and flexibility. It allows for huge volumes of memory to be made available, far beyond the standard current limitation of 48 Terabytes of RAM.