NSP NEWS

(Seoul=NSP NEWS) = SK Hynix has started mass production of the next-generation memory module SOCAMM2 192GB product based on 10-nanometer class 6th generation(1c) LPDDR5X DRAM.

It has strengthened its product supply base to meet the growing demand for low-power, high-bandwidth memory in the AI ​​server market. SOCAMM2 is an AI server-optimized module derived from low-power mobile memory and adapted for server environments.

SK Hynix applied the 1c nanometer process to this product. The company explained that it has achieved more than double the bandwidth and over 75% improved energy efficiency compared to existing RDIMMs.

This product is designed for NVIDIA’s next-generation AI platform, Vera Rubin. It focuses on reducing memory bottlenecks that occur during the training and inference of ultra-large AI models and increasing overall system processing speed.

The demand for memory for AI servers is also changing. As the AI ​​market shifts from a learning-centric approach to an inference-centric one, there is a growing need for memory that offers both power efficiency and high performance.

SK Hynix announced that it stabilized its mass production system early to meet the demand of global cloud service providers.