Nvidia adjusts hbm4 memory supply strategy to address shortages

According to reports from ZDNET Korea, NVIDIA has instructed its upstream HBM memory suppliers to adjust their supply plans for HBM4 memory, increasing the proportion of 8-layer stacked (8Hi) products while reducing the share of 12-layer stacked (12Hi) HBM4. This shift is aimed at addressing the current global shortage of HBM memory, as lower stacking allows for the production of more HBM4 DRAM dies with the same amount of material, thereby increasing the number of Rubin GPUs that can be delivered. The 12Hi stacking configuration is associated with higher heat generation and thermal density, which poses greater challenges for the HBM4E variant, particularly due to the high number of I/O operations it involves. By opting for 8Hi HBM, NVIDIA aims to reduce the rate of DRAM die loss during the manufacturing process. In addition, recent reports suggest that NVIDIA is considering reducing the HBM memory configuration for the Rubin Ultra GPU, likely as part of a broader strategy to mitigate supply constraints. This potential downgrade could involve not only the 12Hi HBM4E variant but also other configurations. The adjustments come amid ongoing discussions about the impact of HBM shortages on GPU production, with multiple sources indicating that NVIDIA is actively exploring ways to optimize its memory supply chain and meet demand without compromising performance.