On the heels of the recent Gen-Z interconnect announcement, an aggregate of some of the most recognizable names in the tech industry have once again banded together. This time, it’s an effort towards the implementation of a fast, coherent and widely compatible interconnect technology that will pave the way towards tighter integration of ever-more heterogeneous systems.
Technology leaders AMD, Dell EMC, Google, Hewlett Packard Enterprise, IBM, Mellanox Technologies, Micron, NVIDIA and Xilinx announced the new open standard to appropriate fanfare, considering the promises of an up-to 10x performance uplift in datacenter server environments, thus accelerating big-data, machine learning, analytics, and other emerging workloads. The interconnect promises to provide a high-speed pathway towards tighter integration between different types of technology currently making up the heterogeneous server computing’s needs, ranging through fixed-purpose accelerators, current and future system memory subsistems, and coherent storage and network controllers.
OpenCAPI sets a new standard for the industry, providing a high bandwidth, low latency open interface design specification built to minimize the complexity of high-performance accelerator design. Capable of 25 Gbits/s data rate, OpenCAPI outperforms the current PCIe specification (which offers a maximum data transfer rate of 16 Gbits/s). Along with the bandwidth improvements, this interconnect alleviates latency issues found in currently-employed communication buses between components and the CPU. With advancement and performance improvement in CPUs hitting a technological wall (and the untimely death of Moore’s Law), big-data and cloud-computing needs have been increasingly serviced by higher-performing solutions, either through Application-Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs, which have seen an increasing demand on the back of search-alghoritms and neural networks processing, prompting giant Intel’s Altera acquisition), or parallel-workload champions GPUs.
Servers and related products based on the new standard are expected in the second half of 2017, on the back of IBM’s POWER9-based servers.
Source: OpenCAPI
