Samsung HBM3 Certified for NVIDIA's China H20 Chip

On July 24, it was announced that Samsung Electronics' latest fourth-generation High Bandwidth Memory (HBM3) has been approved by NVIDIA for use in their processors. This marks the inaugural application of HBM3 in NVIDIA’s technology, with the memory initially allocated exclusively for NVIDIA's H20 AI chips. The H20 is specifically engineered for the Chinese market in response to U.S. policy constraints.

Current information suggests that NVIDIA’s plans for integrating Samsung's HBM3 into additional AI processors remain uncertain. It is also unclear whether further testing will be required before HBM3 can be utilized in other processor models. Samsung has not yet achieved the performance benchmarks for NVIDIA's next-generation HBM3E chips, and ongoing tests are still in progress.

Representatives from NVIDIA and Samsung have declined to comment on this development.

High Bandwidth Memory (HBM) is a type of dynamic random-access memory (DRAM) introduced in 2013. It is designed with vertically stacked memory chips to optimize space and reduce power consumption. HBM is a pivotal component in AI GPUs, essential for processing the vast amounts of data generated by advanced applications.

Sources indicate that Samsung may commence supplying HBM3 for NVIDIA’s H20 processors as early as August. The H20 is the most advanced of three GPUs developed by NVIDIA specifically for the Chinese market, offering notable enhancements in computational capability compared to the H100. Despite a slow initial market reception in China, the H20 is now experiencing substantial growth.