BlockBeats News, August 28th, according to Seoul Economic Daily, Samsung Electronics is developing an 8-layer HBM4E product at Nvidia's request, intended for Nvidia's custom high-bandwidth memory NVHBM. This solution reduces the stacking layers from Samsung's originally planned 12-layer and 16-layer products. Nvidia's target speed is set at 17 to 18Gbps, about a 20% increase from Samsung's early HBM4E samples at 14.4Gbps.

The adoption of an 8-layer design can reduce wafer thinning, precise stacking, and back-end packaging difficulties, alleviating yield pressure and improving supply capabilities. NVHBM may be applied starting from the expected launch of the Rubin Ultra AI GPU next year. Rubin Ultra's GPU interconnect scale will expand from the current 72 cores to up to 576 cores to enable more high-speed GPU cooperation to compensate for the decrease in single GPU storage capacity.

Samsung possesses design and production capabilities for DRAM, logic chips, and basic wafers, providing one-stop service for customized HBM. Industry insiders believe that Samsung has demonstrated higher transmission speeds on HBM4 and, as Nvidia shifts its competitive focus from high stacking to high-speed transmission, Samsung's related capabilities may receive more attention.