SK hynix has begun shipping samples of a new high-bandwidth memory chip aimed at the next wave of artificial-intelligence hardware, according to a report from igor´sLAB carried by Google News.
The new part, called HBM4E, packs 48 gigabytes of capacity per stack. According to igor´sLAB, it runs at 16 gigabits per second per pin — a measure of how fast data moves across each connection — and offers improved thermal characteristics, meaning it is designed to run cooler or shed heat more efficiently than earlier memory.
Shipping "samples" is an early step rather than mass production. It means SK hynix is handing pre-release chips to customers and partners so they can test and design the memory into their own products, in this case the AI chips igor´sLAB describes as "upcoming."
High-bandwidth memory, or HBM, has become one of the most sought-after components in the AI boom. The large processors that train and run AI models need to feed enormous amounts of data to their computing cores, and conventional memory can become a bottleneck. HBM solves this by stacking memory chips vertically and wiring them for very high throughput, which is why capacity, per-pin speed, and heat management are the figures that matter most.
Why it matters: faster, denser, cooler-running memory is a key ingredient in building more powerful AI accelerators, and SK hynix's early HBM4E samples signal where the next generation of AI hardware is heading.