SK Hynix has showcased its most advanced high-bandwidth memory, packaged as a 16-high 48 GB HBM4, designed for high-capacity AI and HPC accelerators that require lots of memory. This is the first time SK Hynix has shown a memory module more advanced than the 12-high 36 GB HBM4, which used to run at 11.7 Gbps. For the new 16-high HBM4, we don't have official data or any speed references, but we can assume SK Hynix managed to extract a bit more bandwidth from the four additional DRAM layers. Interestingly, the announcement was somewhat vague, possibly because Micron and Samsung are also enhancing their HBM4 modules before they become available to companies like AMD and NVIDIA, allowing for potential increases in speed if needed.

The company also demonstrated the custom base die HBM called cHBM. This includes a customized base die sitting at the bottom of the DRAM stack that includes some logic usually associated with the GPU/ASIC logic die, and not memory. For example, this could include HBM PHY and memory controllers that are usually embedded on the GPU die and occupy space, which could be used for more compute logic on the die and maximize performance. In SK Hynix's demonstration and illustration, this is manifested as die-to-die PHYs, embedded memory controllers, HBM PHY, and much more. Customers can basically choose whatever they want and can even embed processing logic inside the die area.