NVIDIA Phases Out First-Generation SOCAMM, Focuses on Next-Gen SOCAMM2 Memory Modules
NVIDIA has reportedly discontinued its first-generation Small Outline Compression Attached Memory Modules (SOCAMM), marking a significant shift in its memory technology strategy. According to industry sources, the company is now prioritizing the development and adoption of SOCAMM2, a next-generation memory module standard designed to deliver higher speeds, greater bandwidth, and improved overall performance.
Advancements in SOCAMM2 Technology
The original SOCAMM modules were developed independently by NVIDIA, outside of established memory standards organizations such as JEDEC. These modules introduced a detachable design and offered impressive technical specifications, including 694 I/O ports—surpassing the 644 ports of LPCAMM and the 260 ports found in traditional DRAM modules. SOCAMM also featured direct integration of LPDDR5X memory and a compact form factor, setting a new benchmark for high-performance memory solutions.
With SOCAMM2, NVIDIA is aiming to further enhance memory performance. The new standard is expected to maintain a similar I/O port count, with 649 ports exposed to the motherboard, but will integrate faster LPDDR5X memory operating at 9600 MT/s, up from the 8533 MT/s of the first generation. There is also potential for SOCAMM2 to support LPDDR6 in the future, broadening compatibility and performance options.
Industry Collaboration and Standardization
One of the key goals for SOCAMM2 is broader industry adoption. NVIDIA is reportedly seeking collaboration with JEDEC to help standardize the SOCAMM2 format, which could pave the way for wider acceptance across the technology ecosystem. Standardization would not only enhance compatibility but also encourage more manufacturers to adopt the new memory module design.
Performance Gains and Supplier Diversification
NVIDIA’s current SOCAMM modules are deployed in the "Blackwell Ultra" GB300 NVL72 system, delivering 18 TB of LPDDR5X memory with a bandwidth of 14.3 TB/s. The transition to SOCAMM2 is expected to boost bandwidth to 16 TB/s, thanks to the faster 9600 MT/s LPDDR5X memory, and could be implemented as a straightforward drop-in upgrade. This performance leap is poised to benefit data centers and high-performance computing applications that demand rapid memory access and large memory capacities.
In addition to technical improvements, NVIDIA appears to be diversifying its supplier base for SOCAMM2. While the first-generation modules were exclusively supplied by Micron, the next generation may see contributions from Samsung and SK Hynix. This move could enhance supply chain resilience and foster competitive innovation, although specific details and timelines have yet to be disclosed.
Looking Ahead
NVIDIA’s transition from SOCAMM to SOCAMM2 underscores the company’s commitment to advancing memory technology for high-performance computing. By pushing for higher speeds, greater bandwidth, and industry standardization, NVIDIA is positioning SOCAMM2 as a leading solution for next-generation memory modules in data centers and beyond.
