NVIDIA plans to manufacture between 600,000 and 800,000 units of its new SOCAMM memory solution throughout this year. The semiconductor company aims to deploy these modular memory components across its artificial intelligence product lineup. Korean publication ETNews reported the production targets for the upgradeable memory technology. The SOCAMM modules represent a departure from traditional HBM and LPDDR5X memory configurations. NVIDIA showcased this technology during its recent GTC conference presentation.
Micron developed the SOCAMM memory using LPDDR DRAM technology typically found in mobile devices and low-power applications. The modular design allows users to replace memory components using three screws rather than permanent soldering methods. NVIDIA's GB300 Blackwell platform serves as the initial deployment vehicle for this memory architecture. The company expects production volumes to increase significantly next year with SOCAMM 2 availability. Memory partners have historically shipped larger quantities of HBM modules to NVIDIA during 2025.
The SOCAMM format delivers bandwidth performance ranging from 150 to 250 gigabytes per second while maintaining superior power efficiency compared to RDIMM solutions. AI personal computers and server systems can benefit from the swappable memory configuration for future upgrades. Samsung and SK Hynix have entered discussions with NVIDIA regarding potential manufacturing partnerships for these memory modules. The technology offers enhanced performance characteristics compared to LPCAMM and LPDDR5X alternatives commonly used in mobile platforms. Industry observers expect SOCAMM adoption to become standard practice for low-power artificial intelligence devices.
Micron developed the SOCAMM memory using LPDDR DRAM technology typically found in mobile devices and low-power applications. The modular design allows users to replace memory components using three screws rather than permanent soldering methods. NVIDIA's GB300 Blackwell platform serves as the initial deployment vehicle for this memory architecture. The company expects production volumes to increase significantly next year with SOCAMM 2 availability. Memory partners have historically shipped larger quantities of HBM modules to NVIDIA during 2025.
The SOCAMM format delivers bandwidth performance ranging from 150 to 250 gigabytes per second while maintaining superior power efficiency compared to RDIMM solutions. AI personal computers and server systems can benefit from the swappable memory configuration for future upgrades. Samsung and SK Hynix have entered discussions with NVIDIA regarding potential manufacturing partnerships for these memory modules. The technology offers enhanced performance characteristics compared to LPCAMM and LPDDR5X alternatives commonly used in mobile platforms. Industry observers expect SOCAMM adoption to become standard practice for low-power artificial intelligence devices.
