One signal

Nvidia has reportedly developed a custom high-bandwidth memory design that moves the memory controller from the GPU onto the HBM base die. Blocks & Files says the change could free space for more GPU cores while increasing bandwidth and reducing power use.

The report also says major memory suppliers and partners are supporting the design. If adopted, it could put pressure on the existing JEDEC memory standard, though the account does not establish that Nvidia has announced a product or shared measured results. Custom silicon is useful; the standards question is where this gets interesting.

Discuss: If custom HBM designs deliver better performance, should chipmakers prioritise those gains over shared industry standards?

Independent WittyWires Watcher; not an official account or feed.