| Mobile Web

Search results for HBM4

Industry

Supermicro starts shipping Nvidia Vera Rubin NVL72 rack, supplying units with 1,152 GPUs

Supermicro said on Sept. 29 it has started shipping Nvidia\'s Vera Rubin NVL72 rack that integrates its data center building block solution DCBBS and direct liquid cooling technology stack DLC-2. The company said it will take orders and supply the rack in scalable units. Each unit supports 1,152 GPUs and 331TB of HBM4 memory. It also said it tests and validates racks with the liquid cooling system integrated to shorten deployment time.

Industry

Samsung foundry narrows losses with HBM4; profit turnaround still unclear

Samsung Electronics\' foundry business is expected to narrow losses in the third quarter as HBM memory volumes lift base-die output made on its logic process. The shift to HBM4 increases foundry-linked revenue within memory sales, and Samsung is also expanding turnkey deals that bundle memory, foundry and packaging for external customers. Kiwoom Securities forecasts the foundry and System LSI operating loss will shrink this year, while Samsung has said a return to profit could be possible soon.

Industry

Samsung Electronics bets on zHBM by stacking memory on accelerators

Samsung Electronics is preparing a two-track strategy of conventional HBM5 and its zHBM concept as it seeks to regain leadership in high-bandwidth memory. The company is shifting to a design that stacks HBM directly on top of AI accelerators, removing the interposer that links logic and memory side-by-side. Samsung targets 230 percent higher bandwidth and 70 percent lower power than HBM4E. Commercialisation will hinge on heat management and wafer hybrid-bonding yields.