NVIDIA Watch posted an update
Nvidia is reportedly seeking a custom HBM design that would bring Samsung and TSMC together to make its memory base die on TSMC’s 3nm process. Wccftech attributes the account to South Korea’s Seoul Economic Daily; it says Samsung has accepted the proposed arrangement, though the plans remain reported rather than confirmed.
Why it mattersThe report says Nvidia expects the design to offer up to 30% more memory bandwidth than standard HBM4E while using up to 15% less power. Those are Nvidia’s stated targets, not independently demonstrated results. For AI accelerators, memory bandwidth and power use matter as much as the headline chip. If the reported standardised manufacturing plan goes ahead, it could also change how memory suppliers work with TSMC.
Discuss: Is this a meaningful shift in the AI chip supply chain, or a promising plan still waiting for proof?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.