Micron has announced the release of its HBM3E 12-Hi memory, boasting an impressive 36GB capacity and a bandwidth of 1.2TB/s. This powerful memory is designed to fuel the demands of artificial intelligence (AI) GPUs, enabling the processing of larger AI models and faster insights.
SK hynix is set to begin mass production of its advanced 12-layer HBM3E memory chips for AI applications by the end of September, with shipments expected to start in Q4 2024. The company is also pushing forward with its next-generation HBM4 memory, slated for release in the second half of 2025, designed to be compatible with NVIDIA’s upcoming Rubin R100 AI GPU.
NVIDIA’s Blackwell Ultra AI GPU is getting a spec bump with up to 288GB of HBM3E memory and a 50% performance increase. However, the rollout has been hit with delays, and a redesigned B200A version is in the works. The news comes from SemiAnalysis, which has also revealed that Blackwell Ultra will come in both standard CoWoS-L and a new MGX NVL 36 form factor.
SK hynix will be attending FMS 2024, a global semiconductor memory event, where it will showcase advancements in its memory technologies and products, including its next-generation AI memory products like the 12-layer HBM3E and 321-high NAND. The company will also present its vision for the AI space and highlight its leadership in the industry.
