NVIDIA Reportedly Shifts to 8-Layer HBM4 to Boost Rubin GPU Production
NVIDIA has reportedly requested its High Bandwidth Memory (HBM) suppliers to increase the proportion of 8-layer (8Hi) HBM4 memory over 12-layer (12Hi) configurations. This adjustment is intended to optimize the supply chain and production capacity for its upcoming Rubin GPUs. By shifting to 8-layer stacks, NVIDIA aims to mitigate severe HBM shortages and accelerate the delivery of Rubin GPUs. Additionally, this move helps address the thermal density and yield issues associated with more complex 12-layer packaging. Producing 8Hi HBM4 requires fewer DRAM dies per stack, allowing manufacturers to output more finished memory units and reduce processing waste. Reports also suggest NVIDIA is considering downgrading the memory specifications of its high-end Rubin Ultra GPU due to these supply constraints.
## BACKGROUND
High Bandwidth Memory (HBM) is a specialized 3D-stacked DRAM technology designed to provide ultra-fast data transfer rates for AI and high-performance computing. HBM4 is the next-generation standard that features significantly increased I/O routing, which makes vertical stacking more complex. NVIDIA's upcoming Rubin GPU architecture, the successor to Blackwell, is designed to utilize this advanced memory technology to power next-generation AI factories.