Nvidia May Cut HBM Capacity Amid Memory Shortage for Rubin Ultra

Nvidia is considering reducing the capacity of high-bandwidth memory connected to its next-generation graphics processing units, according to foreign media reports. The move is interpreted as a response to lowering specifications, as the memory supply crunch makes it difficult to secure sufficient quantities of the latest HBM in time for shipment. The Information, a U.S. information technology specialist outlet, reported on August 6, citing sources, that Nvidia is considering equipping its Rubin Ultra with less HBM capacity than originally planned. HBM is memory that stacks multiple layers of DRAM to increase bandwidth. Rubin Ultra is the next-generation GPU that will succeed Rubin, which is scheduled for release in the second half of this year, and is expected to enter mass production after the end of next year. Nvidia CEO Jensen Huang first unveiled Rubin Ultra at a developer conference last year, stating that each GPU would be equipped with 1TB of HBM4E. However, the samples currently being tested have fewer stack layers and less memory capacity per die. In some samples, memory capacity was limited to 192GB or 256GB, far short of the 1TB presented at last year announcement. Andrew Bell, Nvidia senior vice president of hardware engineering, said last month that the company was in a stable state in terms of memory supply, but in reality it is struggling to secure HBM.