SK Hynix has introduced one of the first computational memory solutions (CMS) of the industry which is a 512GB memory expansion module having to compute capabilities and also a PCIe 4.0 x4 interface with CXL protocol on top.  

The device is to increase performance by several times when compared to CPUs in a few particular computations. 

SK Hynix’s CMS has four 128GB conventional DDR memory modules and also has a system-on-chip that can handle machine learning KNN i.e., k-nearest neighbors classification algorithm acceleration, data filtering, and load balancing featuring a PCIe 4.0 x4 interface which supports both CXL.mem and CXL.io protocols. The CMS card also consists of two network connectors to connect to other nodes. 

The level of power consumption related to the card is still not known, but it has an eight-pin auxiliary PCIe power connector that delivers up to 150W of power. 

 SK Hynix says its prototype CMS can perform data filtering as well as real-time data processing operations more rapidly than many-core CPUs in its big data analysis platform designed together with SK Telecom.  

SK Hynix did not speak about the advantages of the computational memory solution over traditional processors in the press release. But, one of the company’s photos shows graphs and numbers stating the benefits provided by the device. 

Park Kyoung, Head of Memory System Research at SK Hynix stated, with the help of internalization of computational functions, CMS has enabled performance many times faster than dozens of CPU cores in specific computations. Considering this to be a prototype, they think they can bring improvement in the performance even more and are also thinking to apply the technology to other applications such as big data. 

The CXL CMS prototype that has been designed by SK Hynix was demonstrated first at the Open Compute Project (OCP) Global Summit. The company also showed its big data analysis platform optimized along with SK Telecom.