SK Hynix HBM3E memory is now in mass production

Recently, SK Hynix announced the commencement of mass production of its latest memory product, HBM3E, and plans to supply it to key clients by the end of March. These clients include NVIDIA, which possesses GPUs based on the Hopper architecture, namely the H200, and those based on the BlackWell architecture, the B200.

SK Hynix stands as the inaugural supplier of HBM3E, having completed the development of this memory in August 2023. Last month, the company dispatched new 12-layer stacked HBM3E samples to NVIDIA for testing. The HBM3E memory announced for mass production today is presumably an 8-layer stacked product, boasting a capacity of 24GB. SK Hynix asserts that its latest HBM3E product sets the industry benchmark in terms of speed and thermal management. The cutting-edge HBM3E can process 1.18TB of data per second, equivalent to streaming over 230 full HD movies (each 5GB) in a single second. Furthermore, thanks to the adoption of the Advanced MR-MUF process, the HBM3E’s heat dissipation performance has improved by 10% compared to its predecessor, reducing the likelihood of overheating.

Sungsoo Ryu, the head of SK Hynix’s HBM business, stated that the mass production of HBM3E refines the company’s lineup of AI memory products. He expressed the hope that the success of the HBM product line would further enhance clients’ confidence in the company, solidify relationships, and cement SK Hynix’s leading position in the AI memory supply sector.

Previous reports have indicated that in response to the increasingly competitive memory market, SK Hynix had planned to construct a wafer fabrication plant in Indiana, USA, and invest $1 billion in South Korea to build advanced HBM packaging facilities. The initiation of mass production of HBM3E products undeniably secures SK Hynix a strategic advantage in the future memory market.