- Joined
- May 6, 2019
- Messages
- 12,595
- Points
- 113
Could next-generation NVIDIA GPUs feature multiple chiplets? That's beginning to sound like a solid possibility. Chief scientist Bill Dally and SVP of GPU engineering Jonah Alben were recently quizzed about monolithic vs. chiplet designs and hinted they already have the technology and know-how to release such a product.
They aren't expected to come anytime soon, however. Alben admitted that process shrinks haven't reached a point yet where GPU chiplets are actually worthwhile. Dally adds: “This gives us a bunch of technologies on the shelf that at some point in time, if it became economically the right thing to do to, assemble GPUs from multiple chiplets, we basically have de-risked the technology. Now it’s a tool in the toolbox for a GPU designer.”
It is beginning to look as though the days of the big monolithic processor design, whether CPU or GPU, are numbered. Much like the days of processor past, where the beefy single core designs gave way to more lithe multi-core CPUs, the sorts of massive, transistor-heavy chips that have dominated the computing world are increasingly giving way to silicon packages mixing multiple smaller chiplets.
They aren't expected to come anytime soon, however. Alben admitted that process shrinks haven't reached a point yet where GPU chiplets are actually worthwhile. Dally adds: “This gives us a bunch of technologies on the shelf that at some point in time, if it became economically the right thing to do to, assemble GPUs from multiple chiplets, we basically have de-risked the technology. Now it’s a tool in the toolbox for a GPU designer.”
It is beginning to look as though the days of the big monolithic processor design, whether CPU or GPU, are numbered. Much like the days of processor past, where the beefy single core designs gave way to more lithe multi-core CPUs, the sorts of massive, transistor-heavy chips that have dominated the computing world are increasingly giving way to silicon packages mixing multiple smaller chiplets.