SK Hynix Begins Mass Production of 192 GB SOCAMM2 Memory With 2x Bandwidth, a Vital Piece for NVIDIA Vera Rubin

SK Hynix has begun mass production of 192 GB SOCAMM2 memory modules for NVIDIA’s next-gen AI servers.

The LPDDR5X-based module is designed to improve bandwidth and power efficiency in AI data centers.

NVIDIA plans to use the new memory in its Vera Rubin platform, with Samsung and Micron also supplying modules.

The upgrade should help support larger LLMs and other more demanding AI workloads.