perspectiva aiqbenchCipuri & infrastructură

HBM4 is the high-bandwidth memory that feeds AI accelerators, so this timeline matters for anyone tracking AI training and inference capacity. The report suggests 8-layer stacks will ramp for Nvidia's Rubin Ultra-based rack systems around mid-next-year, while 12-layer parts stay the main focus this year. Since these are unconfirmed media reports, the supply schedule is worth watching against Nvidia's own roadmap and shipment data.

Surse care acoperă36Kr, MIT Technology Review, Wired, ZDNet

Samsung, SK Hynix to Begin Mass Production and Deliver HBM4 to Nvidia Next Year

36Kr26 octombrie 20261 min de cititArticolul și perspectiva sunt în engleză.

Media reports say Samsung Electronics and SK Hynix have set the timing for mass production and shipments of 8-layer HBM4 supplied to Nvidia for around the second quarter of next year. A source familiar with the matter said: "As I understand it, Nvidia will adopt 8-layer HBM4 alongside its NVL rack-scale systems based on Rubin Ultra. While 12-layer HBM4 will remain the main focus this year, the share of 8-layer HBM4 is expected to rise significantly from the middle of next year." (Caixin)

Citește articolul complet pe 36Kr

Citește pe 36Kr ↗Întreabă despre acest articol