00:00As AI advances, the chip industry's worries are growing, too.
00:04AI must process ever more data, but the memory that can be attached to a GPU has its limits.
00:11Today, SK Hynix offered a new solution.
00:14Let multiple servers share one giant memory space.
00:18It's called a memory pool.
00:20Instead of endlessly adding memory to each server, they share it as needed to cut data processing waste.
00:26In a test using this approach, the time it took AI to start its first answer dropped from 80 seconds
00:32to 15.
00:34SK Hynix is also working with Intel to validate the technology, going beyond HBM to tackle the limits of AI
00:41systems itself.
00:42Do you think this effort can really reshape the limits of the AI industry?
