Samsung đang phát triển HBM4E 8 lớp cho NVHBM của NVIDIA

The Huoxing Finance report states that on August 28, according to the Seoul Economic Daily, Samsung Electronics is developing an 8-layer HBM4E product at NVIDIA's request for use in NVIDIA's custom high-bandwidth memory, NVHBM. This design reduces the number of stacked layers compared to Samsung’s original plans for 12-layer and 16-layer products. NVIDIA has set a target speed of 17 to 18 Gbps, approximately 20% higher than the 14.4 Gbps of Samsung’s early HBM4E samples. The 8-layer design lowers the complexity of wafer thinning, precise stacking, and backend packaging, easing yield pressures and enhancing supply capacity. NVHBM is expected to be first implemented in the Rubin Ultra AI GPU, scheduled for release next year. The GPU interconnect scale of Rubin Ultra will expand from the current 72 to up to 576 GPUs, compensating for reduced per-GPU memory capacity with more high-speed GPUs working in tandem. Samsung possesses the design and manufacturing capabilities for DRAM, logic chips, and base dies, enabling it to offer a one-stop solution for custom HBM. Industry insiders believe that Samsung has already demonstrated high transmission speeds in HBM4; with NVIDIA shifting its competitive focus from high stacking to high-speed transmission, Samsung’s related capabilities may receive increased attention.