NVIDIA Phases Out SOCAMM Design for Its SOCAMM2 JEDEC Standardized for High Performance Memory

Posted on
NVIDIA transitions from SOCAMM to SOCAMM2, enhancing memory speeds and bandwidth for high-performance computing platforms.

Quick Report

NVIDIA has officially transitioned from its first-generation SOCAMM (Small Outline Compression Attached Memory Module) to the second-generation SOCAMM2. This move aims to enhance memory speeds and bandwidth, making SOCAMM2 a more robust solution for high-performance computing platforms such as AI servers and edge computing systems.

The SOCAMM2 standard, developed with JEDEC's assistance, retains the 649 I/O ports of its predecessor but integrates faster LPDDR5X memory running at 9600 MT/s, compared to SOCAMM's 8533 MT/s. This upgrade is expected to increase bandwidth from 14.3 TB/s to 16 TB/s in systems like NVIDIA's Blackwell Ultra GB300 NVL72. Additionally, SOCAMM2 may support LPDDR6 in the future, with memory sourced from multiple suppliers, including Samsung and SK Hynix.

The first-generation SOCAMM was a proprietary memory module standard developed by NVIDIA, featuring a detachable design and superior specifications compared to traditional DRAM solutions. With 694 I/O ports and direct LPDDR5X memory substrate integration, SOCAMM was tailored for high-performance computing platforms, including AI servers and embedded systems.

SOCAMM2 builds on this foundation by offering higher memory speeds and improved bandwidth. The new standard maintains the same 649 I/O ports exposed to the motherboard but integrates faster LPDDR5X memory running at 9600 MT/s. This represents a significant performance boost over the first-generation SOCAMM, which operated at 8533 MT/s.

One of the key advancements with SOCAMM2 is its potential for broader adoption. NVIDIA is collaborating with JEDEC to standardize the format, making it more accessible for a wider range of applications. This move could position SOCAMM2 as a leading memory module standard for high-performance computing.

Another notable aspect of SOCAMM2 is supplier diversification. While the first-generation SOCAMM was exclusively supplied by Micron, SOCAMM2 may involve multiple suppliers, including Samsung and SK Hynix. This diversification could enhance supply chain resilience and reduce dependency on a single supplier.

In terms of practical application, SOCAMM2 is designed to be a simple drop-in replacement for existing systems. For example, NVIDIA's Blackwell Ultra GB300 NVL72 system, which currently ships with SOCAMM, could see its bandwidth increase from 14.3 TB/s to 16 TB/s with the adoption of SOCAMM2. This upgrade underscores the performance benefits of the new standard without requiring significant system redesigns.

Written using GitHub Copilot GPT-5 mini in agentic mode instructed to follow current codebase style and conventions for writing articles.

Source(s)

  • TPU
  • ET News