SK Hynix announced two months ago that it had developed the first CXL memory sample based on DDR5 DRAM and planned to put it into mass production in 2023. Now SK Hynix has taken it a step further and announced that it has successfully developed the industry's first CXL storage solution CMS (CMS) equipped with computing functions, integrating computing functions in a 512GB CXL memory module.
At the recent OCP (Open Compute Project) Global Summit held in San Jose, the United States, SK Hynix unveiled CMS and its software and hardware platform for the first time, showing customers the next generation of high-performance computing solutions and their value. It is expected that CMS module will be installed on the next generation server platform to improve system performance and energy efficiency.
Open Phoenix News to view more HD pictures
As an open interconnection protocol, CXL based on PCIe has higher bandwidth. It can enable high-speed and efficient interconnection between CPU and GPU, FPGA or other accelerators, meet the requirements of today's high-performance heterogeneous computing, and provide higher bandwidth and better memory consistency. SK Hynix, based on the CXL protocol, integrates scalable high capacity memory and computing functions, and introduces the CMS module.
The CMS module of SK Hynix uses four 128GB traditional memories and a system on a chip, providing machine learning and data filtering calculation functions commonly used in big data analysis and other applications. It has PCIe 4.0 interface, supports CXL.mem and CXL.io protocols, and has two network connectors for connecting to other nodes. The power consumption of the CMS module is uncertain for the time being. It is equipped with an 8Pin external power supply interface, which can provide 150W power.
SK Hynix plans to make relevant presentations at the SK Technology Summit held in South Korea in early November. In the future, it will actively carry out research and development and cooperation related to CXL memory to provide customers with different added values.
