SK Hynix paves the way to next-gen AI by completing HBM4 development

SK Hynix is the first company that’s ready to mass-produce HBM4 memory

SK Hynix has announced that it has completed its development of HBM4 (High Bandwidth Memory 4) memory and that they are prepared to start mass production. This new memory type was designed to deliver the next generation of high-performance AI. With boosted memory bandwidth and efficiency, HBM4 is ready to solve the data bottleneck that limits today’s AI datacenters.

With 2,048 I/O terminals, double that of HBM3, HMB4 can deliver vastly more data throughput than today’s memory solutions. SK Hynix has claimed that its HBM4 can deliver 40% more power efficiency, and can “improve AI service by up to 69% when this product is applied”.

SK Hynix has also claimed that its new HBM memory has exceeded the expectations of JEDEC. The JEDEC standard operating speed of HBM4 is 8 Gbps. SK Hynix has delivered speeds of over 10 Gbps.

(Images from SK Hynix)

To create its new HBM memory, SK Hynix has used its 5th-generation 10-nanometer technology. This minimises its risks for mass production, ensuring that the company can create its new HBM4 memory modules in high volumes. Note that SK Hynix plans to continue improving its HBM4 technology over time.

High-performance AI hardware is often bandwidth-constrained. If hardware manufacturers can adopt HBM4 memory quickly, they will be able to deliver significant performance gains with their next-generation hardware, even without other advancements.

You can join the discussion on SK Hynix’s completion of its HBM4 memory development on the OC3D Forums.