SK hynix Presents AI Memory Technology Vision at Supercomputing 2026
SK hynix showcased its technology vision centered on memory-centric solutions at Supercomputing 2026 held November 15-20 in Chicago, Illinois. The company demonstrated its MARS (Memory-centric AI Rack System), CXL pooled memory, Compute-using-DRAM, and Custom HBM Array technologies aimed at improving AI infrastructure efficiency. SK hynix also highlighted its product lineup including HBM4E, HBM4, HBM3E, server DRAM modules, and eSSD solutions for next-generation AI applications.
Key takeaways
SK hynix presented MARS (Memory-centric AI Rack System) demonstrating pooled memory sharing KV cache at rack level, AI agent serving with computational memory pools, and token-sharing technology based on CXL pooled memory.
The company showcased KV Cache Offloading over CXL, which adds a CXL memory pool to the AI memory hierarchy alongside HBM and DRAM, verified with SK Telecom in a real LLM inference environment.
Compute-using-DRAM technology performs computations inside memory using DRAM cell charge-sharing behavior, enabling logic operations and arithmetic directly in memory to reduce data movement.
Custom HBM Array connects multiple cHBM units with computational functions to overcome conventional GPU limitations on HBM unit connections and improve LLM decoding efficiency.
Product lineup displayed included HBM4E and HBM4 alongside HBM3E, server DRAM products (256GB 3DS DDR5 RDIMM, 128GB DDR5 MRDIMM, 96GB DDRIMM, 192GB SOCAMM2), and eSSD solutions (PEB210 E1.S with DLC support, PS1110 E3.S).
Technical Leader Younghoon Min presented on enabling agentic AI serving with disaggregated memory pools, showing GPU and CPU utilization improvements over conventional methods.
The exhibition attracted approximately 10,000 visitors and 500 companies and institutions.
Primary source
TickerStar generated this brief from an official company, investor-relations, earnings-call, or regulatory source. Review the primary material before making an investment decision.
Open Investor relationsFollow SKHY in real time
Get material earnings, filing, IR, and earnings-call alerts by email.