SK Hynix Launches Mass Production of 192 GB SOCAMM2 Memory for NVIDIA’s Vera Rubin With Double the Bandwidth

SK Hynix Launches SOCAMM2 Memory for Next-Gen AI Data Centers

SK Hynix has officially started mass production of its innovative SOCAMM2 memory, designed to meet the increasing demands of NVIDIA and next-generation AI data centers. This cutting-edge memory module offers capacities of up to 192 GB, addressing the significant computational needs of modern AI applications.

Three Solidigm SSDs labeled 'SOCAMM2' are displayed on a white background.

Enhanced Performance and Efficiency

According to We Hynix, the SOCAMM2, built on a 1cnm technology process, boasts more than double the data transfer bandwidth and features over 75% enhanced power efficiency compared to traditional RDIMM (Registered Dual In-line Memory Module) solutions. These improvements make the SOCAMM2 particularly well-suited for high-performance AI workloads.

Targeting NVIDIA’s Vera Rubin Platform

The SOCAMM2 products are specifically engineered to work seamlessly with the NVIDIA Vera Rubin platform. This collaboration reinforces We Hynix’s commitment to becoming a leading provider of memory solutions optimized for artificial intelligence applications.

“We will solidify our position as the most trusted AI memory solution provider, through close collaboration with our global AI customers.”

SK Hynix anticipates that the introduction of SOCAMM2 will significantly alleviate memory bottlenecks often faced during the training and inference stages of large language models (LLMs), which can contain hundreds of billions of parameters. This advancement is expected to play a critical role in speeding up the overall processing capabilities of AI systems.

Shifting Market Focus

With a noticeable shift in the AI market from inference to training, the SOCAMM2 memory module is garnering attention as a next-generation memory solution, capable of efficiently managing LLMs with reduced power consumption.

The image shows an SK Hynix 'SOCAMM2' memory module next to a gold-colored heat spreader.

SOCAMM2 represents a significant evolution in memory technology, adapting low-power memory—previously utilized in mobile devices—to server environments. This module is poised to become a primary memory solution for the next wave of AI servers.

SOCAMM2 (Small Outline Compression Attached Memory Module 2) is an AI server-optimized memory module based on LPDDR technology, offering a compact form factor, high scalability, and a compression connector that enhances signal integrity while facilitating easy module replacement.

Source & Images