Myron Xie joins @JordanNanos on why NVDA's Rubin Ultra preview had 1TB of HBM per package vs the 192GB that ships.
It comes down to supply and not performance, why 4-HI more than doubles the cubes you can harvest per wafer, why the loudest calls for less memory are coming from the labs themselves, and when the shortage finally eases.
The gap between Nvidia's 1TB HBM preview and the 192GB that ships comes down to wafer yield from 4-HI die stacking, not a performance tradeoff, explaining current memory scarcity and giving a signal for when the shortage might ease.
