I wouldn't call it HBM as a concept an abject failure, as the tech has found a niche in Nvidia and AMD's top dog price-is-no-object accelerators.

Problem was that AMD tried to sell cards with the tech to consumers, which don't really benefit from the high bandwidth part of the tech, so all it really did at the end of the day was bump up the cost and limit VRAM amounts.

Also, tbd on new info from RDNA3, as its supposed to include MCM.