Nvidia Unveils Custom HBM Design With 30% More Bandwidth Than HBM4E

Delivers 30% higher performance than HBM4E

Market for custom chips poised to expand

Nvidia has unveiled NVHBM, a next-generation high-bandwidth memory technology. The design is a form of custom HBM that alters the standard HBM configuration made by Samsung Electronics and SK Hynix to strengthen the computing capabilities of artificial intelligence accelerators.

Nvidia disclosed NVHBM in a company blog post on Aug. 26 alongside its second-quarter earnings report. The technology centers on boosting computing performance by relocating the memory controller used in chips such as graphics processing units, or GPUs. The memory controller governs what data a GPU pulls from HBM and how it retrieves it during processing.

In conventional designs, the memory controller sits inside the compute chip. In Nvidia's NVHBM design, it is moved to the base die, the bottom layer of the stacked DRAM in HBM. That opens up space for additional computing circuitry and lifts performance.

Nvidia said chips built on the NVHBM design would improve memory performance and power efficiency. Memory bandwidth would rise 30% from HBM4E, the seventh generation of HBM. That means more data can move between HBM and the compute chip each second. Power consumption could fall by as much as 15%, while the area available for computing functions could increase by up to 25%.

Nvidia does not appear set to manufacture NVHBM itself. The company said it is building a standard NVHBM implementation that will be offered through multiple memory suppliers. It also said it would jointly develop NVHBM with Annapurna Labs, Amazon's semiconductor design subsidiary. That suggests Nvidia could ask memory chipmakers such as Samsung Electronics and SK Hynix, which supply it with HBM, to manufacture the product.

The move is fueling expectations that demand for custom HBM will increase. HBM suppliers such as Samsung Electronics and SK Hynix have so far produced memory based on standardized specifications. But as AI infrastructure evolves, big tech customers are seeking different AI chip specifications. An industry official said customers led by Nvidia will increasingly demand HBM tailored to their own requirements.

Samsung Electronics' custom HBM, or cHBM, unveiled at the Hot Chips 2026 semiconductor conference in the US on Aug. 23, was built on the same concept. Most memory companies expect next-generation HBM to be developed primarily as custom products, another industry official said.

Roh Yu-jung, Hankyung.com reporter yjroh@hankyung.com