great question - i would imagine only minimal architecture tweaks in current applications ( not replacing HBM4 ) - memory transfer rates related to AI is the increasing bottleneck - this chokes compute performance significantly - traditionally when that bottleneck is addressed increased power issues appear and heat becomes a major issue....this 4DS som solves both the compute performance issue, the power and heat issue within existing architectures ( if successful at 20nm of course)
happy to be corrected
4DS - Anything but Charting, page-30852
-
- There are more pages in this discussion • 1 more message in this thread...
You’re viewing a single post only. To view the entire thread just sign in or Join Now (FREE)
