SOCAMM3: New RAM Standard for NVIDIA Rosa and AMD Florence
The SOCAMM (Small Outline Compression Attached Memory Module) standard is rapidly establishing itself in the industry. While the first two generations are gaining traction, development is already underway for the third iteration: SOCAMM3. Designed for energy-efficient DRAM chips, this standard is aimed at the AI and server markets starting in 2028, specifically targeting NVIDIA Rosa and AMD Florence platforms.
Technical Specifications of SOCAMM
SOCAMM builds upon the foundations laid by CAMM and CAMM2, focusing on an ultra-low profile and enhanced power efficiency. Current module specifications describe a compact form factor: 90 mm in length and 14 mm in width. Unlike vertical UDIMMs, these modules are laid flat and secured with three screws to ensure optimal contact across the entire length. The design is single-sided; all memory chips are mounted on the top surface, while the underside houses 694 I/O contacts.
The Roadmap for SOCAMM3 and LPDDR6
According to reports from South Korean media, TLB—a manufacturer of SOCAMM PCBs—is currently developing SOCAMM3 and is reportedly in discussions with two major customers. While final specifications have not been officially confirmed, mass production is anticipated to begin in 2028. It is expected that SOCAMM3 will utilize the LPDDR6 standard, providing the high bandwidth and low power consumption required for intensive AI workloads.
This shift is expected to impact major hardware players. For NVIDIA, SOCAMM is a natural fit for the Rosa processor line, intended as a successor to Grace and Vera. For AMD, the transition appears inevitable; the company has already confirmed that its future server CPUs will support "Next Gen Memory." This suggests that the Zen 7 (Florence) and Zen 8 (Ravenna) architectures, slated for late 2028, will likely incorporate SOCAMM to meet the memory demands of modern data centers.
Гика
