Qualcomm and AMD are reportedly considering the integration of SOCAMM2 memory into their artificial intelligence product lines to enhance memory capacity and speed. Following NVIDIA’s ‘Vera’ CPU, which features LPDDR5X memory on the SOCAMM factor, offering 1.2 TB/s of memory bandwidth and supporting up to 1.5 TB of LPDDR5X memory, Qualcomm and AMD aim to pursue a similar strategy to achieve high speed and high capacity artificial intelligence systems. These systems require fast memory placed physically close to artificial intelligence accelerators, and SOCAMM is positioned as a complement to the extensive HBM memory that already resides on many accelerators.
In this architecture, SOCAMM acts as a fast memory pool with sufficient capacity to store entire models directly in memory, which reduces the need for frequent data transfers to SSD flash storage. NVIDIA’s SOCAMM approach involves surrounding the CPU with multiple SOCAMM modules, and the report suggests that Qualcomm and AMD are expected to adopt a similar layout in their own artificial intelligence focused solutions. By using modules around the processor rather than fixed soldered memory, vendors can bring bandwidth and capacity closer to compute without redesigning every board.
AMD’s Instinct MI accelerators, when paired with EPYC CPUs, are highlighted as likely candidates for this SOCAMM2 integration, although AMD may also opt to create a completely new system design to take full advantage of the form factor. Qualcomm already offers artificial intelligence solutions like the AI200 and AI250 inference accelerators, which include up to 768 GB of LPDDR5 per card, and these products could soon be complemented by SOCAMM. The article explains that SOCAMM would allow standardized memory expansion based on the SOCAMM standard rather than soldering individual memory modules, enabling manufacturers to offer different system versions and configurations simply by adding or removing SOCAMM modules, and to avoid extensive PCB soldering work when scaling memory.
