Between adding a computing core to HBM and the zHBM (stacking HBM packages) that they talk about in the article I can't imagine the yield for this setup is going to be any better than what they're doing now.
How are they doing with building more fabs... because they're going to need them.
This has potential to be huge in the long run as latency is one of the major bottle necks to get AI to the next level. Should be interesting to see how this plays out.
12 Comments
Mr_Dragonspears@reddit
howaboutbecause@reddit
Vb_33@reddit
ML7777777@reddit
floorshitter69@reddit
Blueberryburntpie@reddit
liliputwarrior@reddit
Ghostsonplanets@reddit
Big-Newspaper646@reddit
tempest_@reddit
ThankGodImBipolar@reddit
Big-Newspaper646@reddit