Nvidia Unveils NVHBM High-Bandwidth Memory Architecture for AI Cards
Nvidia introduced a customized High Bandwidth Memory architecture, NVHBM, claiming significantly higher bandwidth, lower power consumption, and more usable silicon area than conventional HBM4E. Nvidia says it moves the memory controller into the HBM base die, reducing XPU interface circuitry and freeing up to 25% more compute die space, positioned as part of NVLink Fusion for custom AI silicon.