SK Hynix announced today that it has begun mass production of 192GB SOCAMM2 memory modules. This product is based on 1cnm process (sixth generation 10nm technology) LPDDR5X low-power DRAM, which is the next generation memory module standard.

SOCAMM2 is a new module that adapts low-power memory originally used in mobile products such as smartphones to the server environment. It is designed as a main memory solution for next-generation artificial intelligence servers.

It features a slim profile, high scalability, and compression connectors for improved signal integrity and easy module replacement.

Compared with traditional RDIMM,SK Hynix emphasized that the bandwidth of the mass-produced version of SOCAMM2 has more than doubled and the energy efficiency has been optimized by more than 75%, providing an optimized solution for high-performance AI computing.

RDIMMs are DRAM modules commonly used in servers and workstations that contain registers or buffer chips that relay address and command signals between the memory controller and the DRAM chip.

SK Hynix specifically pointed out that its SOCAMM2 product is specially designed for the NVIDIA Vera-Rubin platform.

SK Hynix expects that this new product will fundamentally solve the memory bottleneck encountered in training and inference of large language models with hundreds of billions of parameters, thereby significantly accelerating the processing speed of the overall system.

As the focus of the AI ​​market shifts from inference to training, SOCAMM2, which can run large language models with low power consumption, is gaining widespread attention.

To meet the needs of global cloud service provider (CSP) customers, SK hynix has stabilized its mass production system in advance and provides a complete product portfolio.

Justin Kim, President and Head of AI Infrastructure at SK Hynix, said that with the launch of 192GB SOCAMM2, the company has set a new standard for AI memory performance and will solidify its position as the most trusted AI memory solution provider by working closely with AI customers around the world.