The GPU giant is set to produce a decent inventory for its modular memory solution (SOCAMM) for AI products, making it easier to upgrade memory on the devices for superior performance and high efficiency.
NVIDIA is Reportedly Getting Hundreds of Thousands of LPDDR-Based SOCAMM Memory for AI PCs; Demand to Boom with Next-Gen SOCAM 2 Product
The new SOCAMM memory, which was showcased at the NVIDIA GTC event, is set to see an increase in inventory this year as NVIDIA aims for superior performance and lower power consumption on its AI products. At the GTC event, we saw NVIDIA's GB300 platform using the SOCAMM memory, developed by Micron, which is quite different from the HBM and LPDDR5X memories used on AI products such as servers and mobile platforms.
The SOCAMM memory is based on the LPDDR DRAM, which is traditionally used in the mobile and low-power devices, but unlike solutions like HBM and LPDDR5X, the SOCAMM is upgradable. It's not soldered on the PCB and can be secured via three screws. As per the latest report by the Korean publication ETNews, NVIDIA is set to produce about 600,000-800,000 units of the SOCAMM memory this year. This increase in production of the new memory is set for deployment in NVIDIA's AI products stack.
One of the first products with SOCAMM memory is the latest GB300 Blackwell platform, which already hinted at NVIDIA's intention to transition to a new memory form-factor for many of its AI products. While the target of up to 800,000 units this year is far lower than the HBM memory shipped to NVIDIA by its memory partners in 2025, it will begin to scale up next year, particularly with the availability of the SOCAMM 2 memory.
The SOCAMM memory brings a custom form-factor, which is not just compact and modular but is also significantly more power-efficient than RDIMM. We don't know the exact power-efficiency improvements but we do know from reports that SOCAMM will deliver superior efficiency while also bringing much higher bandwidth than RDIMM as well as LPDDR5X and LPCAMM, which are one of the most popular memory solutions for mobile platforms.
With about 150-250 GB/s of memory bandwidth and being swappable, SOCAMM offers a versatile option for AI PCs and AI Servers, which can be upgraded easily. In the near future, it's expected to see SOCAMM becoming the norm for low-power AI devices and while Micron is the current manufacturer for SOCAMM for NVIDIA, Samsung, and SK Hynix are supposedly in talks with NVIDIA to manufacture the modules for the company as well.
About the author: Sarfraz Khan is a hardware reporter with a focus on PC components and the builder community. With years of experience writing about PC hardware and laptops, his work has been featured on several reputable technology publications. Sarfraz's hands-on experience is demonstrated through his first-person accounts of using and comparing different hardware configurations, providing practical and relatable insights for everyday users. His technical analysis is respected by peers in the enthusiast community and has been cited by specialized hardware sites such as Germany's Igor's Lab.
Follow Wccftech on Google to get more of our news coverage in your feeds.
