- CMS is industry’s first CXL-based memory solution with computational function
- SK hynix, SK Telecom collaborate on joint development of integrated platform in hardware-software for computing system optimization
Just two months after introducing the first CXL memory sample in August, SK hynix has successfully developed a computational memory solution, becoming the industry’s first to integrate computational functions into CXL memory. The solution, a result of a close collaboration with SK Telecom from early on for optimization of the next-generation, high-performance computing infra system, marks a meaningful case where SK ICT companies converged their technologies for a joint development.
Unveiling both the CMS and its software platform at the Open Compute Project (OCP) Global Summit 2022 taking place from Oct. 18 in San Jose, USA, SK hynix showed a next-generation high-performance computing infrastructure solution as well as the value of the solution from the customer’s perspective. The solution is expected to be installed on next-generation server platforms to improve system performance as well as energy efficiency.
Integrated Solution for Memory, Computational Functions Introduced
CXL (Compute eXpress Link)*, a base technology for the latest development, is a new interconnect technology designed to promote efficiency of various solutions such as GPU, AI accelerator and memory.
Among the advantages of CXL that have caught the attention of the industry is its flexible increase in memory capacity. CXL enables the flexible expansion of a system’s memory capacity by installing a memory card like a GPU or SSD. SK hynix focused on CXL’s technological characteristic of supporting both memory and accelerator to proceed with a preemptive research and development and, consequently, come up with the latest integrated solution. Park Kyoung, Head of Memory System Research at SK hynix, noted, “CXL has created an opportunity for memory providers.”
The CMS technology provides not only the advantage of CXL but also showcases advantages of machine learning and data filtering computation functions frequently performed by big data analysis applications. The solution also screens operations that are inefficient for the CPU to perform or that consume a lot of energy during the transfer of data between the memory and the CPU. Park Kyoung, announcing follow-up research and development plans, said, “Through the internalization of computational functions, CMS enabled performance several times faster than that of dozens of CPU cores in specific computations. Considering that this is just a prototype, we think we can improve the performance even further and are considering applying the technology to other applications such as big data.”
Convergence of SK companies’ technologies
Park also said, “When introducing new technologies and their solutions to the market, it is important to prove their value from the customer’s point of view, but the possible scope of collaboration with customers is relatively limited, especially for data-center server areas. In that sense, collaboration with SK Telecom provided a great opportunity to bring the CMS into reality. SK Telecom’s experience in AI/Big Data-based service infrastructure development and operations was reflected in each part of the development process.” In order to efficiently process large amounts of data, SK Telecom has developed and ran in various commercial services its own Lightning DB*, an in-memory data analytics platform.
The two companies proceeded with development of both CMS and CMS-applied Lightning DB simultaneously. Through this process, SK hynix was able to discover and apply the computational functions required by customers, and SK Telecom was able to enhance the competitiveness of its platform.
Yang Seung-ji, Vice President and Head of Vision R&D at SK Telecom, said, “It is common for hardware solutions of a new concept to take a considerable amount of time to prove their effectiveness from a software application’s point of view. But this time, meaningful computations for SK Telecom’s actual application services were selected, which helped us save a significant amount of time by jointly performing all processes from structure design of hardware-software to development and verification.” Yang continued, “As we verified the performance improvements of the solutions, we plan to apply them to screening tasks that increase accuracy in large-scale AI learning data in the future and use them to strengthen the competitiveness of SK Telecom’s AI service.”
SK ICT affiliates have been operating R&D projects for synergies to converge various technologies of related companies that encompass semiconductors, hardware solutions, software platforms, and AI services. Lee Jong-min, Vice President and Head of Future R&D at SK Telecom, said, “This case is significant in that we successfully developed a technology by combining AI and semiconductor capabilities of SK Telecom and SK hynix together. We will continue to provide new technology-based value to customers and contribute to the ICT R&D ecosystem globally.”
The CXL memory solution prototype developed by SK hynix and the big data analysis platform optimized together with SK Telecom will be unveiled initially at the Open Compute Project (OCP) Global Summit and will also be demonstrated at the SK Tech Summit in Korea in early November. In the future, SK hynix plans to actively conduct R&D and collaborations related to CXL memory that provides added value from the customer’s point of view.