tech
SK hynix and NVIDIA announced a multiyear partnership to co-develop next-generation memory, with SK hynix set to supply chips for NVIDIA's Vera Rubin AI supercomputers and Jetson Thor robotics platforms.
HBM4, or high-bandwidth memory, is at the center of the NVIDIA-SK hynix partnership. HBM4 works by stacking multiple DRAM memory chips vertically and connecting them with extremely short, wide pathways, letting data flow to a GPU far faster than memory chips laid out flat on a circuit board, which matters enormously for training and running today's largest AI models.
DDR6 is a next-generation memory standard aimed at general computers, not specialized AI accelerators, and GDDR7 is designed mainly for graphics cards handling video game rendering. Neither is built for the extreme bandwidth demands of massive AI training hardware the way HBM4 is.
Memory bandwidth, not raw processing power, is often the real bottleneck in AI hardware today, since a chip's calculations are useless if data can't be fed to it fast enough, which is exactly the problem stacked memory like HBM is designed to solve.
tech
What was notable about this arrangement?In public-key cryptography, what are the two keys called that a user has?What type of memory chip do most solid-state drives use to store data persistently?What is this Google quantum chip called?Which BMW model did these humanoid robots help assemble?What is this standard called?Which company has been rapidly expanding driverless robotaxi service across U.S. cities in 2025-2026?True to its title, what key technology did the Transformer boldly leave behind?What did the reward drop to after the fourth halving in 2024?What is this IBM processor called?What model did OpenAI launch in August 2025?As of 2024, approximately what share of the AI training chip market does NVIDIA hold?Quration — Quration Play